Semantic Technology

1
Semantic Technology Healthcare in BT

• John Davies
• Head of Next Generation Web BT Research

2
Overview

• Introduction to BT and UK Healthcare IT activity
• Relevance of Semantic Technologies to healthcare
  IT integration standards (SNOMED-CT HL7)
• Implementing SNOMED-CT (large structured
  vocabulary) in OWL
• Further opportunities and obstacles for adoption
  of Semantic Web technologies

3
Part 1 BT, Semantic Web and Healthcare

• Next Generation Web Research Group
• Coordinates
• Core partner
• BTs aim is to translate this research into
  benefits for customers in divisions such as BT
  Global Services

4
BT Global Services

• Provides networked IT solutions for multi-site
  organisations (gt30 total BT revenue)
• The Health division has a prominent role in
  10bn UK NHS National Programme for IT
• 3bn of contracts for
• Networking
• National application service provider
• Local application service provider for London

5
Health IT is heterogeneous distributed
Hospital ward
Community health
Pathology
Pharmacy
General practice
Surgery
Radiology
6
Pattern repeated in each health community
7
Patient centric Health IT
Patient journey
8
NHS Connecting for Health

• National Care Records Service
• A live, interactive patient record service
  accessible by health professionals whether they
  work in hospital, primary care or community
  services
• The core of the NHS CRS is provided by the Spine
• Central summary record
• Detailed patient information stays local
• records of detailed history of care,medical
  conditions, tests, X-rays scans and other
  results.
• ? A large distributed system containing
  semantically heterogeneous information that must
  be integrated to some degree around individual
  patient care, with aggregated data sent to the
  Spine

9
Tools for Healthcare IT Integration

• SNOMED-CT - Consistent use of medical vocabulary
• Systematised Nomenclature Of Medicine - Clinical
  Terms
• Merger of UK and US efforts with a joint
  editorial board
• Original development College of American
  Pathologists
• Now managed by an international organisation
  based in Denmark
• Large reference terminology (gt400,000 concepts)
• Aim to be machine interpretable for the exchange,
  aggregation, analysis of clinical data and to
  enable clinical decision support
• ? semantic technology!
• HL7 v3 - Consistent messaging between
  applications
• Specifications for consistent exchange of
  information between clinical applications
• SNOMED-CT provides a vocabulary whilst HL7
  provides the data model with which messages are
  built

10
Patient centric Health IT
HL7 messages containing SNOMED-CT terms
11
What is the Semantic Web?
"The Semantic Web is an extension of the current
web in which information is given well-defined
meaning, better enabling computers and people to
work in cooperation." -- Tim Berners-Lee, James
Hendler, Ora Lassila, The Semantic Web,
Scientific American, May 2001
Today's Web A place where computers do the
presentation (easy) and people do the linking and
interpreting (hard). Why not get computers to
do more of the hard work? XML, RDF, OWL
12
Health the Semantic WebSNOMED-CT OWL

• SNOMED-CT
• Concept based
• Clear separation of lexical representation and
  conceptualisation
• Whipples procedure and pancreatoduodenectomy
  terms represent the same medical concept
• Cold can mean cold sensation or a common cold

• OWL
• Directly corresponds with OWL Class
• Class(Common_cold)
• RDFS label can be used to represent term labels
  (synonyms)

13
Pure subsumption hierarchy

• SNOMED-CT
• has an is-a relationship equivalent to logical
  implication / subsumption
• A patient with asthma implies a patient with a
  respiratory disorder

• OWL
• Corresponds to OWL subclass
• SubClassOf(Asthma Respiratory_disorder)

14
Property based definitions

• SNOMED-CT
• Relationships with other concepts that partially
  or fully define the concept of interest
• Description logic reasoner used to classify
  concepts

• OWL
• Majority can be regarded as existential
  restrictions
• Description logic reasoner used to classify
  classes

An appendicectomy is defined as a surgical
procedure using the method excision and the site
appendix structure Class(Appendicectomy defined
intersectionOf(Surgical_procedure
restriction(method someValuesFrom Excision
restriction(procedure_site someValuesFrom
Appendix_structure))
15
Post coordination

• SNOMED-CT
• Not designed to be complete out of the box
• Extensible at the point of data entry through
  post coordination

• OWL
• Corresponds to anonymous class expressions of
  OWL-DL

Concepts for kidney excision and left exist,
together with rules that specify it is
appropriate to combine them together. intersectio
nOf(Excision restriction(procedure-site
someValuesFrom intersectionOf(kidney
restriction(laterality someValuesFrom left))))
16
Implementing SNOMED-CTSearching for the right
term

• gt400,000 concepts
• Clinician may realistically expect to enter 2-15
  terms per encounter (based on Primary care
  experience)
• Adding terms should not increase documentation
  time.
• Term search boxes most efficient (for now)

17
Key search features
Unordered start of word search within phrase

• Responsive users will not have the time or
  willingness to page through multiple pages of
  results

18
Recording using SNOMED-CTBrowsing for the right
term

• What happens when you cant find the right term?
• Guidance is to use the next most specific term
  available and append information as free text
• Users need to know specific versus general terms
• Users need to browse the hierarchy

19
Browsing features

• Navigating top down in SNOMED-CT not
  straightforward (too large)

20
Browsing features

• Localised browsing on search hits

21
Browsing features

• Localised browsing down on search hits

22
Browsing features

• Localised browsing up on search hits

23
Terminology reasoning at the point of care

• Description Logic reasoning is required in the
  live environment
• Why?
• Post coordination (i.e. allowing users to build
  anonymous class expressions)
• If user has built allergy almond how does the
  application work out to retrieve this information
  when someone else asks for patients with allergy
  to nut?

24
1 Clinician records clinical data using novel
expression e.g. Allergycaused_byalmond
Clinician A
Healthcare application
25
Clinician A
2. Application registers novel expression with
terminology service
Healthcare application
Terminology service
26
3. Terminology service uses DL reasoner to
calculate novel expressions position in the
subsumption hierarchy of existing concepts
Clinician A
Healthcare application
Terminology service
Description Logic Reasoner
27
Clinician A
Healthcare application
Terminology service
Description Logic Reasoner
4. Terminology service provides information to
application on the subsumption hierarchy
including the position of the novel expression
e.g. allergycaused_byalmond is-a nut allergy
28
Clinician A
Healthcare application
Terminology service
Description Logic Reasoner
Clinician B
5 Clinician queries for patient data which
contains novel expressions e.g. What patients
have nut allergies?
29
Clinician A
Healthcare application
Terminology service
Description Logic Reasoner
6 Application uses hierarchy information provided
by the terminology service to execute query.
(Almond allergies are now known to be nut
allergies).
Clinician B
30
Terminology Services example 2
Snomed CT
Data Warehouse
Reasoner
Clinician App
Admin App
Pre-coordination
Post-coordination
Q How many Key hole procedures?
Hernia Repair
Access Instrument?
Laparoscope (implies keyhole procedure)
31
Progress in BT

• Proof of concept using Semantic technology
• FaCT Description logic reasoner from University
  of Manchester
• 30 mins to initially load over 400,000 concepts
  in SNOMED-CT
• lt10ms to calculate subsumption of simple
  anonymous class expressions shown in example
• Also evaluating a range of other reasoners

32
Other opportunities

• Natural Language Generation
• rendering the formal SNOMED-CT structure into
  natural language
• Systems interoperability
• Integration of heterogeneous data sources
• Data warehouse mining
• Mapping ontologies (terminology models)
• (Semi-)automatic completion of EPR from
  clinicians free text

33
NLG - Summarisation

• NLG takes structured data in a knowledge base or
  ontology and generates natural language text
• Motivation
• natural presentation of ontology fragment for
  end-user
• automatic documentation of ontologies
• generate textual reports from formal knowledge
• Keeps texts constantly up-to-date reflecting
  changes in the ontology
• SEKT technology OntoSum (University of
  Sheffield)

34
Summary Structuring

• Capture regular patterns can be applied
  recursively
• Describe-Instance -gt Describe-Parents,
• Describe-Attributes, Describe-Part-Whole,
  ...
• Aggregation
• Lexicalisation
• e.g. hasName -gt is called

35
Description of HSBC
Financial Institution
Person
Organisation
Bank
lendsTo
lendsTo
HSBC
employees
market-cap
43bn
137000
36
Description of HSBC
37
Systems Interoperability today
multiple data transformations
38
Semantic Technology-based
39
Information integration
Applications
Semantic bus
No need to re-engineer legacy data sources
Courtesy Tim Berners-Lee and W3C
40
Semantic Web for HealthcareGartner June 2006

• Potential to revolutionize healthcare
  interoperability
• Enable sharing of knowledge, information and
  rules
• Ontologies and rules are not new concepts in
  healthcare informatics SNOMED-CT
• Commoditize decision support currently considered
  leading edge in CPR systems
• Unify information and decision processes across
  admin, clinical, research and strategic boundaries

41
Issues in adoption of Semantic technology in
Healthcare

• Ontology management
• Although formalism of SNOMED-CT and OWL very
  similar they are not identical
• We need to assess the implications and agree with
  customers that any differences are acceptable
• Todays ontology environments have difficulty
  with the scale of SNOMED-CT
• Often designed for schema ontologies of 100
  classes not 400,000 classes

42
Other issues

• RDF (Resource Description Framework) is
  promising
• a clear mechanism for the identification of
  resources (URI Universal Resource Identifier)
• a mechanism for the aggregation of data from
  distributed sources.
• reification which allows statements about
  statements.
• a link to the well defined semantics of OWL.
• But..
• Novelty exposure to RDF is limited in this
  community
• Alternative technology Infrastructure and tools
  are currently wedded to relational data model
• Scalability and performance flexibility of RDF
  comes with the downside of reduced performance.
  Although examples are appearing of RDF
  repositories containing millions of RDF
  statements, more evidence of performance and
  scalability will be needed to ensure its adoption
  (this is improving continually)
• Hybrid solutions may allow a compromise

43
Is Semantic Technology ready for Healthcare?

• Yes
• certainly in ontology management and reasoning
• a very natural fit
• clinicians are used to ontologies
  (terminologies)
• No
• Perceived as not quite yet ready for deployment
  in other areas such as data management?
• Maybe
• Is Healthcare ready for Semantic Technology?
• need Healthcare IT customers to agree

44

• Thank you for your attention
• Questions?