TRANSFoRm: Vision of a learning healthcare system

1
TRANSFoRm Vision of a learning healthcare system

• Vasa Curcin, Imperial College London
• Theo Arvanitis, University of Birmingham
• Derek Corrigan, Royal College of Surgeons Ireland
• www.transformproject.eu
• TRANSFoRm is partially funded by the European
  Commission  - DG INFSO (FP7 247787)

2
TRANSFoRm Consortium
3
Knowledge in healthcare

• Clinical trials
• Controlled populations
• Well-defined questions

• Distilled scientific findings
• Usable in clinical practice
• Decision support

• EHR systems
• Wide coverage
• Vast quantity
• May lack in detail and quality

4
Aims of TRANSFoRm

• To develop the infrastructure in primary care to
  support
• Epidemiological research using GP records,
  including genotype-phenotype studies and other
  record linkages
• Research workflow embedded in the EHR
• Decision support for diagnosis
• Infrastructure components
• Implementation methods
• Models for data and process organization
• Services
• Architectures
• Demonstrations with industry

3
5
TRANSFoRm Use Cases
GORD
Type 2 Diabetes
Decision Support

• Research Question
• Effectiveness of continuous versus on demand PPI
  use?
• Electronic CRF embedded in the eHR
• RCT with event-initiated patient-related outcome
  measures
• Trigger within EHR
• Semantic Mediator
• eCRF tool (embedded in EHR)

• Research Question
• Are well selected SNPs in T2D patients associated
  with variations in drug response to oral
  antidiabetics?
• Genotype-Phenotype record linkage study
• Privacy model
• Record linkage (browsing, selecting, extracting)
• Data quality tool
• Provenance tool

• Experimental Study Comparing approaches to
  diagnostic support
• Alerting versus suggesting Clinical Prediction
  rule web service (with underlying ontology)
• Prototype DSS integrated into EHR system
• Domains
• chest pain
• abdominal pain
• shortness of breath

6
Overall Architecture (1)
7
Overall Architecture (2)
End User Tools and Services
Study Design
Query Workbench
Study Management
GP Decision Support
Eligibility Criteria Designer
Identify/Recruit Eligible Patients
Interactive Consultation Decision Support Tool
Protocol Designer
Consent Data Manager
eCRF Designer
Study eCRF Data Collector
Timeline Designer
Query formulation and execution tool
Patient-reported outcome Manager
CDE Designer
Clinical Evidence Service
Data Mining and Analysis
Data Provenance
Semantic Mediation
Vocabulary Service
Clinical Evidence Extraction Tool
Provenance Capture Service
Provenance Audit Tool
Data quality tool
Support Services (e.g. rule based security,
authentication)
Middleware (Distributed Infrastructure)
Distributed Nodes
8
Models in TRANSFoRm

• Clinical Data Integration Model (CDIM)
• Mapping clinical data from EHRs and aggregated
  data repositories
• Clinical Research Information Model (CRIM)
• Research process information
• Evolution of Primary Care Research Information
  Model (PCROM).

9
9
Provenance

• Processing history of a data item
• Actors involved
• Data sources used
• Operations performed
• Data movements
• Authorizations
• Supports
• Reliability
• Accountability
• Auditability
• Related research areas
• Workflows standardizing representation of data
  processing
• Information systems tracking data evolution

10
Provenance benefits

• System monitoring
• Entity oriented, rather than raw logs
• Exploratory investigations possible
• User accountability
• Data items and actions directly connect back to
  the user
• Traceability
• Every step in the evolution of a result becomes
  easily accessible
• Reuse and repurposing
• Storing previous queries for use by
  recommendation engines and the users directly
• Data warehouse
• Large-scale statistics about resource usage,
  organisational performance and user activity

11
Architecture
12
Example Provenance of a query
13
Example Linkage process
14
Example Decision support
15
The challenge of representing knowledge in an
interoperable computable form

• Developing a user understandable, computable and
  extensible knowledge representation scheme for
  capturing clinical trials concepts and
  information (knowledge)
• including patient safety (medical errors)
• with a multilingual support
• The foundation of interoperability lies with a
  shared understanding of concepts and data
  representation between systems
• it is necessary to establish both syntactic
  (model-based) and semantic interoperability to
  represent knowledge in a computable form
• TRANSFoRm provides this through an Integrated
  Vocabulary Service (TRANSFoRm VS)

16
Challenges for EU TRANSFoRm

• Vocabulary Services have been traditionally
  provided by the NCI
• Cross Mapping of Vocabularies
• The case of Primary Care in Europe the
  difficulty with using the NCI EVS in European
  Primary Care is that neither Read codes nor ICPC2
  are part of the service
• some mapping exists with UMLS (but many concepts
  missing)
• EVS can be augmented by referencing Read codes to
  UMLS codes via SNOMED CT (where a cross-mapping
  exists) and by adding the ICPC-2 mapping through
  ICD-10
• Evolving technologies and APIs for EVS
• LexEVSAPI various versions from 3.0 to 5.0

17
Application Architecture

• The TRANSFoRm Integrated Vocabulary Service is
  designed to allow end users to search and
  retrieve clinical vocabulary concepts and
  associated content
• a web interface and a web service API
• the service uses the LexEVS (version 5.1)
  technology to access a backend UMLS vocabulary
  database
• the service uses direct Java Database
  Connectivity (JDBC) to access other vocabulary
  databases (e.g. Read Codes V2, ICPC2)

18
TRANSFoRm Integrated VS Server Architecture
19
A screenshot of the web-based interface
20
RCD v2/ICPC2

• Read Codes (RCDv2) and International
  Classification of Primary Care (ICPC2) corpus of
  terms and their associated mappings
• created to cater for the initial need of the
  existence of specific primary care oriented
  terminologies.
• The UK NHS Connecting for Health Terminology
  Centre - mappings from Read Codes version 2 to
  SNOMED CT.
• The Read Codes v2 database in Transform VS is set
  up based on this mapping so that Read Codes 2
  concepts can be linked to a UMLS search. Similar
  approach for ICPC2.
• ICPC2-ICD10 Thesaurus and mappings - Transition
  Project _at_ University of Amsterdam
• The TRANSFoRm team is updating the ICPC-ICD 10
  mapping and Thesaurus
• to eventually incorporate the new Thesaurus into
  UMLS.

UMLS Metathesaurus
SNOMED CT Codes
Read Codes v2 Codes
UMLS Metathesaurus
ICD-10 Thesaurus/Codes
ICPC2 Codes
21
Demonstration of integrating a current prototype
of the Study Designer and TRANSFoRm Integrated
Vocabulary Service
22
ePCRN Study Designer and TRANSFoRM VS integration
23
eCRF Introduction

• eCRF is electronic version of case report form
  (CRF)
• CRF are forms designed by clinical study
  investigators to collect data about each subject
  during the whole study process
• CRFs are usually filled in by research staff, but
  can also be filled in by practice staff if CRFs
  are integrated into practice system
• Collected data are saved and analysed afterwards

24
eCRF Workflow
Protocol Timeline CRFs
Study Database
Design Study
Collect Data
Analyse Data
The following provide some example screenshots of
the eCRF technology on a current prototype
25
Protocol Design
26
Timeline Design
27
CRF Design
28
Decision Support Tool

• The Decision Support Tool will provide
  patient-specific advice at the moment of
  consultation so that clinicians are able to
  access and quantify likely differential diagnoses
  framed in terms of diagnostic probability and
  alternative diagnostic possibilities.

• Decision Support Tool Characteristics
• Embedded within the eHR
• Triggered by a reason for encounter
• Presents diagnostic prompts based on ontology
  service for clinical prediction rules
• Collects ontologically controlled diagnostic cue
  data
• Alerts/suggests for potential missed diagnoses

29
Decision support objectives

• Develop a model of clinical evidence upon which
  diagnostic decision support is provided
• Formulating and quantifying potential
  differential diagnoses based on presenting
  patient cues from primary care EHRs
• Linked to primary care EHRs and based on clinical
  knowledge derived from electronic sources of
  research data

30
Method

• Ontological approach - define an ontology of
  clinical evidence to represent what we know about
  defined clinical scenarios
• Provides for top down and bottom up reasoning to
  formulate potential diagnoses
• Clinical Prediction Rule used as one possible
  mechanism to interpret that evidence by applying
  some quantification of the importance of the
  constituent diagnostic cues

31
General model of evidence
32
Development Tools

• Protégé Ontology Development
• Sesame Triple Store provides persistent
  representation enabling dynamic update of
  knowledge
• Sesame API provides for programmatic
  update/manipulation and provision of clinical
  evidence web service for decision support

33
Protégé defining concepts and relations
34
Protégé defining instances of knowledge
35
Sesame query platform
36
Future DSS work

• Focus moving to defining query and update
  interfaces as part of a defined evidence service
• Allow for generation and update of knowledge from
  data mining done on electronic sources of primary
  care data
• Working with EHR vendors to define user interface
  requirements

37
TRANSFoRm Vision

• TRANSFoRm will drive the integration of clinical
  research and clinical practice by developing
  tools services to facilitate greater system
  interoperability.

TRANSFoRm Tools Services
Pharmaceutical Companies
CTDMS
CTDMS
CTDMS
eHR System
eHR System
eHR System
eHR Vendors
Clinical Research Networks
Clinical Repositories
Clinical Repositories
Clinical Repositories
CTDMS Clinical Trial Data Management Software
e.g. Oracle Clinical eHR Electronic Heath Record
e,g, EMIS Web Clinical Repositories e.g. GO-DARTs
Tayside
38
TRANSFoRm Anticipated Benefits
Support quicker and more economic recruitment and
follow-up of Randomised Clinical Trials...
...with an integrated eHR interface that enables
the rich capture of clinical data, including
symptoms and signs
Improve patient safety...
...by providing not only a diagnosis support tool
but also a query workbench that supports the
identification of patient eligible to participate
in clinical trials
Support large scale phenotype-genotype
association studies and follow-up on trials...
... through distributed interoperability of eHR
data and clinical data repositories that maintain
provenance, confidentiality and security
Drive the integration and re-use of clinical data
stored in different eHR systems...
... with software tools and web-services that
support clinical research by enabling use of
controlled vocabulary and standardised data
elements
Enhance uptake of eHR systems that offer support
for clinical care and research...
... by adopting an open-source business model,
allowing eHR vendors and data integrators direct
cost savings and the ability to reach more
customers through improved pricing flexibility
39
Contact details

• www.transformproject.eu
• 1st year deliverables publicly available
• eCRF EHR integration
• Theo Arvanitis, University of Birmingham
• Decision Support System
• Derek Corrigan, Royal College of Surgeons Ireland
• Provenance, general queries
• Vasa Curcin, Imperial College London