Modelling Medical Knowledge and Processes

1
Modelling Medical Knowledge and Processes

• Tony Rose
• Software Development Manager
• Advanced Computation Lab

2
Contents

• CR-UK and the Advanced Computation Lab
• Cognitive science the clinical context
• Clinical guidelines modelling languages
• The Credo Trial
• Usability issues

3
Cancer Research UK

• History
• Europe's largest independent cancer research
  organisation
• 2002 merger of ICRF and CRC
• Vision
• to conquer cancer through world-class research,
  aiming to control the disease within two
  generations
• Funding
• research spend for 2002/2003 191 million
• Structure
• In-house research labs
• External research grants

4
Advanced Computation Lab

• One of CR-UKs in-house labs, based in London
• Mission
• research in the computer and cognitive sciences,
  including AI and cognitive psychology, as a basis
  for developing innovative information
  technologies to support patient care
• Structure
• Scientists developers

5
Research Aims

• To carry out fundamental research in theories of
  computation and intelligent systems
• To develop novel technologies based on
  theoretical results
• To demonstrate the practical value of these
  technologies in biomedical applications for
  clinical research and patient care

6
Research at the ACL
7
The Clinical Context

• Perhaps 16,000 lives could be saved if all
  current knowledge of cancer were properly
  applied
• ICRF Vision for Cancer, 1995
• Up to 97,000 unnecessary deaths p.a. in the US
  are due to medical error. Total national costs
  between 17B and 29B
• Institute of Medicine, To Err is Human 1999

8
The Knowledge Crisis

• Modern medicine is becoming a humanly impossible
  task
• New knowledge is being generated too quickly for
  individuals to absorb and for organisations to
  assimilate into mainstream clinical practice

9
The Role of Clinical Guidelines

• CGs based on best available evidence from
  systematic research
• Following clinical guidelines helps to save lives
• improve care quality and standardize care
• Various sources
• National Institute for Clinical Excellence
• National Guideline Clearinghouse
• many more

10
CGs and CIGs

• Most CGs are deployed in textual form
• Narrative, flowcharts, diagrams
•   BUT
• busy clinicians have little time to read
• memories are unreliable, working pressures acute
• conventional guidelines address general
  principles of care, not the needs of individual
  patients
• Our aim is to provide evidence-based decision
  support in executable form

11
CIG Modelling Languages

• Various paradigms, e.g.
• procedural
• rule-based
• e.g. Arden syntax
• task-based
• GLIF, Eon, Asbru, PROforma

12
PROforma

• Formal language for modelling clinical knowledge
  and processes
• supports the definition of clinical guidelines
  and protocols in terms of
• a set of primitive tasks which are composed into
  networks representing clinical processes
• logical constructs that allow the details of
  tasks and their inter-relationships to be defined

13
Task hierarchy

• Uses Argumentation for decision making
• weigh up arguments for or against a particular
  proposition

14
PROforma

• Plans container structures that hold any number
  of task instances
• Enquiries requests for information or data from
  an external agent (e.g. the end-user)
• Actions typically clinical procedures (such as
  the administration of an injection) to be carried
  out
• Decisions choices of some kind, such as a choice
  of investigation, diagnosis or treatment

15
Tallis

• Software environment for developing and executing
  clinical decision support and workflow
  applications
• Clinical processes are modelled using the Tallis
  Composer
• Processes can be enacted to provide workflow and
  DS via a web browser

16
(No Transcript)
17
(No Transcript)
18
Knowledge Sharing

• Tallis supports knowledge sharing through a
  repository server, e.g. OpenClinical.net
• leverage work done by others
• reduce development costs
• Leads to knowledge publishing lifecycle
• iterative development and review process

19
Formalised in PROforma
Protocols, guidelines, care pathways

Tested in silico (in the lab)
Trials

Feedback into research and policy
Routine use
20
(No Transcript)
21
The Credo Trial

• as a basis for developing innovative
  information technologies to support patient care
  -gt the Credo Trial
• multi-centre trial of integrated software package
  designed to support breast cancer care throughout
  the patient journey
• Collaborators
• Guys Hospital Breast Unit, London
• Addenbrookes Hospital Breast Unit, Cambridge
• Carol Franc Buck Breast Care Centre, UCSF

22
Trial Questions

• What is the baseline quality and safety of
  decision-making in the management of breast
  cancer?
• Can PROforma yield improved consistency, quality
  and safety of clinical decision making through
  the patient journey? How and where?
• Can such services be acceptable to and valued by
  clinicians? Freeing them from administrative and
  operational burdens?
• Can this result in improved experience for the
  patient?

23
(No Transcript)
24
Credo Core Service Model
25
Phase 1

• Total of 222 clinical, patient and communication
  services identified
• about 65 require professional decisions
• Software demonstrator constructed
• A number of prototypes implemented in PROforma
  and web technologies
• Pilot trials initiated

26
Phase 2

• Continue pilot trials
• Initiate clinical trials of specific services
• Develop infrastructure
• Clinical database
• Patient records
• Communications messaging
• Plan multi-centre trial of integrated care
  manager

27
Phase 3

• Phase 3
• RCT of integrated patient journey system
• Quantitative assessment of clinical outcomes
• compliance with pathway
• effectiveness and timeliness of communication
• patient satisfaction

28
Credo Phase 1 Results

• Aim trial of support for initial assessment
  (Triple Assessment) of women with breast symptoms
• Methodsoftware assists clinician in calculation
  of familial risk, selection of appropriate
  investigations and subsequent management of
  patient
• Study design Crossover experiment with balanced
  blocked design, each clinician managing a set of
  10 simulated cases, 5 with and 5 without DS

29
Credo Phase 1 Results

• Aim to evaluate potential effect of PROforma DS
  on clinical decision making in Triple Assessment
• Decision model
• Pathway included 4 decisions familial risk, type
  of imaging, type of biopsy, and management
• Materials and methods
• 15 hypothetical paper cases developed by an
  expert panel of 5 judges (2 breast surgeons, 1
  breast pathologist, 1 radiologist and 1
  geneticist). Optimal management for each case was
  agreed by consensus
• 24 doctors asked to manage 5 cases with and 5
  without DS. Balanced block design used to
  allocate cases

30
Triple Assessment
Hurt C, Patkar V, Steele R, Rose T, Fox J (in
preparation) Steele R, Fox J Proceedings of
European conference on AI in Medicine, 2003
31
Triple Assessment Study Results
24 participants (17 consultants, 5 specialist
registrars, 1 nurse practitioner) Average number
of years in speciality 9.3 (range 1 - 23)
Would patient care improve with TA decision
support? In favour 16 Undecided 1 Against
7 Would they personally wish to work with TA
decision support? In favour 12 Undecided
3 Against 9
32
Credo Phase 1 Results

• Outcome measures deviation from the guidelines
  (as agreed in advance by a panel of experts

33
Usability Issues

• High level (conceptual)
• How best to model medical knowledge?
• What kind of conceptual structures do clinicians
  use?
• What kind of decision making framework?
• How should those concepts be represented by the
  system (UI)?
• Low(er) level
• Does the UI employ HCI best practice?
• Does it support the users task appropriately?

34
The Tallis Usability Trial

• Aim to compare Tallis Composer 1.3 with Tallis
  Composer 1.5
• Hypothesis Tallis 1.5 is more usable for new
  users
• Study design between-groups trial with each
  participant performing 6 tasks with 1.3 or 1.5
• Outcome measures
• For each task
• Classify as successful / unsuccessful / partial
  completion
• Assign score based on each level of completion
• Overall average score across all
  tasks/participants

35
Paper Prototyping

• Tallis 1.3 is a working prototype, BUT -
• Tallis 1.5 is a design proposal it exists on
  paper only
• How do we test a design on paper?
• Use paper prototyping
• paper versions of screens widgets
• one person acts as computer
• both UIs are mocked up
• think-aloud protocol

36
Collaboration

• MEDAL (Prof. Sriram)
• Extensive online repository of medical algorithms
• www.medal.org
• Integrate Tallis Medal
• Ontologies, EMRs, lab IS, etc.
• Create an online, open access repository of
  executable medical knowledge

37
Questions?