A Dynamic Problem-Knowledge Coupling Semantic Web Service

1
A Dynamic Problem-Knowledge Coupling Semantic Web
Service

• M.N.Kamel-Boulos, A.V.Roudsari, and E.R.Carson
• Centre for Measurement and Information in
  MedicineCity University, London, UKE-mail
  M.Nabih-Kamel-Boulos_at_city.ac.uk

2
Background and Introduction - 1

• Many information needs arise during everyday
  clinical practice, including questions related to
  diagnosis, determining risks/prognosis, planning
  an optimum investigative or therapeutic strategy,
  disease prevention and patient education
  material, etc. The quality and outcomes of
  patient care will suffer if these information
  needs are unmet or are answered with inaccurate,
  non-current or misleading information.
• Contextual relevance is an important aspect of
  medical information quality that is frequently
  overlooked. It is essentially an information
  retrieval issue. An information resource that is
  excellent on its own merits can end up becoming
  mere distracting noise if it does not properly
  address the users specific knowledge needs, or
  does not fit the clinical context in which it was
  recalled.

3
Background and Introduction - 2

• Problem-knowledge coupling (PKC) aims at
  providing contextually appropriate medical
  knowledge in the right place and at the right
  time. This should be a major goal when designing
  the Electronic Patient Record (EPR).
• Appleyard and Malet (1997) were among the first
  to mention that the incorporation of
  nomenclatures such as UMLS into meta-tags will
  allow the linking of Web-based knowledge sources
  into electronic medical record systems. The
  ideal online medical library should act as a
  contextual medical knowledge provider within a
  Clinical Information System, coupling
  problem-specific knowledge with real patient data
  (EPR). This can assist in decision making,
  improve patient care and educate the student,
  professional and patient.

4
Background and Introduction - 3

• HealthCyberMap (HCM - http//healthcybermap.semant
  icweb.org) aims at mapping medical Web resources
  in unique and novel ways to deliver a
  semantically superior experience to its users.
  This is achieved through intelligent
  categorisation and interactive hypermedia
  visualisation of the medical cyberspace using
  metadata, clinical codes (a terminology or
  classification) and GIS (Geographic Information
  Systems) technologies.
• A clinical terminology/classification is a kind
  of ontology by definition as it preserves and
  understands the relationships between the
  1,000s of terms in it or else it would become a
  mere dictionary/thesaurus. By tagging a medical
  Web resource (or a metadata record of it) with
  clinical codes, we are automatically establishing
  the relationships (as defined by the coding
  scheme in use) between this resource and other
  related (tagged) resources, and also similarly
  coded EPRs. This is the basis of PKC.

5
Background and Introduction - 4

• The Semantic Web initiative (http//semanticweb.or
  g) aims at creating a Web where information
  semantics are represented in a form
  understandable by machines as well as by
  humans. In the case of medical information, this
  can be achieved using clinical codes. This will
  pave the way to more intelligent
  machine-to-machine communication (e.g., between
  an EPR system and a digital medical library), and
  ultimately empower humans.
• Hendler (2001) provides an extensive discussion
  of Semantic Web Services. He mentions a service
  advertise/ discover mechanism. Microsoft has
  already implemented closely related ideas in its
  .NET/ Web Services framework.
• In the case of HCM, we thought of a very useful
  Semantic Web Service, namely a PKC Service to be
  consumed by EPR clients. Coupling is possible as
  both the EPR and HCM will be using the same
  clinical coding system or two different schemes
  with reliable cross mapping.

6
Materials and Methods - 1

• HCM relies on stand-alone metadata (in a central
  databasecf. peripheral metadata embedded in the
  Web resources themselves) based on the Dublin
  Core (DC) metadata set (http//dublincore.org).
• Clinical codes are used to populate the DC
  subject field in HCM database. The current HCM
  Web implementation uses ICD-9-CM, a hierarchical
  classification with no support for multiple
  parentage, as some of HCM hypermaps rely on a GIS
  extension that only understands ICD-9.

7
Materials and Methods - 2

• Candidate resources are hand-selected. Their
  attributes, including ICD codes representing
  their subjects (using code locators, e.g.,
  http//eICD.com) are manually compiled in HCM
  database.
• Manual indexing ensures the quality of selected
  resources and the precision of their topic
  indexing. HCM currently allows three DC subject
  fields (for topic indexing) per resource record.
• The database is registered on HCM server (a
  Windows 2000 Server) as an ODBC Data Source. ASP
  Server Pages are used to query the database and
  present the results on the Web.

8
Materials and Methods - 3

• HealthCyberMap resource metadata entry form in
  Microsoft Access 97.

9
Results - 1

• In our Web Demo, we assumed an EPR that codes
  diagnoses in ICD-9-CM, and since HCM crisply
  describes the subjects of the Web resources
  stored in its database using the same codes, a
  link between the two systems in the form of
  http//healthcybermap.org/icd.asp?SearchTextPUT_
  SINGLE_ICD-9-CM_CODE_HERE will be all what we
  need to perform the coupling query and
  dynamically link the EPR to contextually relevant
  knowledge and guidelines.
• The ICD code from the EPR is passed as an SQL
  query argument to HCM. Only resources with a DC
  subject field containing the argument are
  retrieved.

10
Results - 2Screenshot of HealthCyberMaps
Problem-Knowledge Coupling Demonstrator
(http//healthcybermap.semanticweb.org/pk.htm).
11
Discussion - 1

• The service can be also implemented as a .NET
  Service (giving clients more control over
  presentation of query results), or according to
  any other Semantic Web Service standard when
  available.
• The philosophy behind Semantic Web Services
  including Microsofts implementation is that
  systems should be able to talk to each other
  instantly and reliably without the headache of
  developing ad hoc information exchange protocols.
  Systems should advertise and provide their
  services in a reusable and flexible form to all
  disparate clients (service consumers) who might
  want to use these services and to integrate them
  in their interfaces.

One service and different clients/interfaces
12
Discussion - 2
HCM
EPR

• HCM should ideally be expanded to include more
  comprehensive clinical coding systems like
  SNOMED-CT (when released).
• Clinical codes provide a common backbone language
  (ontology) for communication between the EPR and
  systems like HCM. Combined with a suitable
  terminology server (Bechhofer, 1997), we can
  reason with these codes in more sophisticated
  semantic ways and support synonyms and related
  concepts (parents, siblings, children, cousins
  and uncles).
• To maximise contextual relevance, indexed
  resources should ideally be qualified using a
  mechanism similar to MeSH qualifiers or
  subheadings, which are used to better define a
  topic, narrow retrieval, or express a certain
  aspect of a main heading (ICD has no similar
  support).

13
Discussion - 3

• Using a terminology server to enhance DC subject
  queries in HealthCyberMap.

14
Discussion - 4

• Unlike other inflexible solutions with hard-coded
  knowledge like Dr. Weeds Problem Knowledge
  Couplers (http//www.pkc.com), HCM solution is
  flexible and dynamic. We can keep adding/
  deleting resources to HCM and have all new
  changes instantly reflected on our output without
  modifying HCM architecture or code (or the
  clients calling code).
• We can choose to call certain resource categories
  we need most, e.g., only official guidelines on a
  given topic, instead of all resource types on
  that topic, thanks to the HCM DC type field.

15
Conclusions - 1

• Our study shows the feasibility of EPR to HCM
  problem-knowledge coupling using ICD codes as
  crisp problem-knowledge couplers or hooks. Though
  ICD-9-CM codes are mainly used for EPR billing
  (US), we have successfully demonstrated another
  clever use of these codes using mainstream
  technology that most interested groups can easily
  adopt.

16
Conclusions - 2

• By minimising irrelevant leads (noise) and
  reducing the time needed to find relevant
  information (the right contextually relevant
  knowledge is coupled with real patient data in
  the EPR), the system is clearly beneficial.
• Empirical evidence also shows that well-informed
  physicians and patients are able to make better
  clinical decisions that positively affect
  healthcare outcomes. However, an evaluation study
  will be needed to accurately measure how the
  system is affecting clinical outcomes. The
  success of the system will depend on the quality
  and granularity of metadata it uses (including
  the clinical coding scheme(s) used for topical
  indexing) and the quality of resources it points
  to.