Guideline Support Tools: Current Research

1
Guideline Support Tools Current Research
Computer-based Support for Clinical Practice
Guidelines and Protocols
K Dube Computer Science Dublin Institute of
Technology
2
Presentation Outline

• EWGLP 2000
• 1st European Workshop on Clinical Guidelines
  Protocols,
• Leipzig, 13-14 November 2000
• Conclusion Research Challenges

3
EWGLP 2000

• Medical Decision Making and CPGs vs Workflow
• PROforma CPG Specification Language
• Therapy Planning vs Workflow Asgaard/Asbru
• Guide-X Approach to CPG Formalisation
• ONCODOC DSS vs Clinicians Intellect
• Dynamic Decision Models for CPG Development
• CPG Representational Formalisms and
  Computational Methods EONs Dharma Model
• CONCLUSION Research Challenges

4
Medical Decision Making CPGs vs
Workflow(Stefanelli M, Pavia)
Efficiency
Quality of Care
Pressure
Healthcare Organisation
Cost
Cost Containment
Evidence-based Medical Decision making (use of
CPGs)
Effective Management of Knowledge
Medical Research
5
CPGs vs Workflow
Activities of a medical team
Behavioural aspects of medical work
CPG
Careflow
Organisational Model
Workflow in the Medical Domain
Specific Healthcare Organisation
6
Technologies for Supporting Careflows (Stefanelli
M)

• production rules, semantic nets, Petri Nets
• objects, frames

Organisational structure, Actors, Roles and
Resources
Workflow Process Definition
Careflow
CPG
Flowchart
Petri Net
Simulation
CfMS
7
PROforma CPG Specification Language (Bury, Fox
and Sutton)
Purpose Building practical systems for CPG-based
task management Description logic programming,
object-oriented, task-based (tasks organised into
generalisation hierarchies) Limitations non-exten
sible generic tasks, no goal definition, no
external interaction (EHCR), unparameterised
tasks Applications prescribing, cancer advice,
pain management, anti-retroviral therapy, DSS in
mammography, prescribing scheduling acute
lymphoblastic leukemia
8
Further Limitations of PROforma

• necessity for using global data representations
• no mechanism to encapsulate data locally inside
  task
• compromise engineering principles loosely
  coupled tasks,
• reusability of tasks
• data dependencies not well supported
  (parameterised tasks)
• one task may need to use data produced by
  others
• method to view tasks data inspecting
  attributes
• need to specify data items invoved in
  dependency
• guideline visualisation is at a single level of
  detail
• at subplan level, other parts of guideline not
  visible
• not possible to get global overview of nested
  guideline
• structure

9
Strengths of PROforma

• temporal constraints
• reasoning about time-points, time-intervals and
  
• time-constraints
• ordering constraints among tasks
• ability to backtrack when chosen path proves
  incorrect or more information become available
• ability to abort when patient recovers
• executable model
• once task attributes are filled, an executable
  protocol becomes available
• no further implementation is required

10
PROforma Task Ontology
11
Therapy Planning vs Workflow Asgaard/Asbru(Miksc
h, Kosara Seyfang)
Therapy Planning vs Workflow Map of the Fields
12
Asgaard/Asbru (Miksch, Kosara Seyfang)
Asbru time-oriented, intention-based,
skeletal-plan specification language CPGs
represented as time-oriented skeletal plans in
Asbru Plan Components preferences, intentions,
conditions, effects, plan body Basic syntactic
construct temporal pattern, i.e. one or more
parameter proposition or plan state description
13
Guide-X Approach to CPG Formalisation (Svatek,
Kroupa Ruzicka, Prague)
CPG in computer tractable format (Formal
Representation)
CPGs in text format (Natural Language)
Medical Domain Ontology
PatMan verify presence of ontological concepts
in text (top-down)
Guide-X convert text into a knowledge model or
ontology (bottom-up)
CPG in Plain Text
14
Guide-X in context
Analysis of compliance
Need to process many EPRs, with limited
interaction
Need to minimise information loss and subjective
bias
Requires a model robust to missing and
untypical data
Requires a transparent, text-centred
formalisation
Guide-X
Two-tiered model
Step-by-step, mark-up-based approach
Use of XML, PMML, GLIF semantics
Standards should be reused
15
Scheme of Guide-X
PMML DTD
XHTML DTD
XHTML
GLKL
OCML
GLML-S
GLML-R
GLKL DTD
GLML-S DTD
GLML-R DTD
16
ONCODOC

• Implementation
• CPG is a Structured Knowledge Base, therapeutic
  expertise
• encoded as a decision tree flowchart
• Use
• The Physician reads and interprets the CPG
  through a hypertext
• navigation of the decision tree
  flowchartperformed in a browser.
• Comparison with other systems
• Other systems automatic triggering of CPGs
  logic based on
• patient data
• ONCODOC categorisation of patient to a formal
  encoded
• equivalent from which CPG recommendation is
  derived
• Evaluation
• SOMPS 80 compliance
• IGR 88 compliance

sharability
17
ONCODOC DSS vs Clinicians Intellect (Seroussi
et al)
18
Dynamic Decision Models for CPG Development(Zhu
and Leong)
Methodology basic idea
19
Methodology Tasks for DDM construction in
DynaMoL
Specify a dynamic decision problem type, its
duration, optimality and evaluation criteria
Define the alternative actions and the states
involved
Impose relevant constraints among the decision
factors when appropriate
20
Conceptual model for DDM
Assumptions
Event variables
Actions
Numerical parameters
States
Value functions
Transition functions
Conditional Probability distributions
Basic Characteristic of decision problem
Constraints
Strategic constraint
Declaratory constraint
21
CPG conceptual model
CPG
Objective
Outcomes considered
definitions
Alternative Actions
Others
Method used
Terms used
Health outcomes
Economic outcomes
22
EONs Dharma CPG Model (Tu and Musen)

• EON Approach aim to build a component-based
  architecture
• for constructing guideline based medical DSS
• Dharma CPG Model task-based, also
  component-based, handles CPG complexity and
  variability by
• core CPG model
• extensible set of tasks, alternative methods
• allowing configuring of model to include only
  necessary tasks and methods in a specialised
  model for use in target CPG applicaton
• Core Model Components clinical algorithms,
  domain model, criteria languages

23
Representation Formalisms - Summary(Tu and Musen)
24
Conclusion Research Challenges

• CPG acceptance and compliance
• medical community awareness, conviction,
  compliance
• informatics improve formal CPG representation
  and execution,
• improve availability
• CPG dissemination and utilisation
• making CPG site-specific
• managing exceptions (where implementation fails
  after
• CPG adaptation) - organisational problems

25
Conclusion Research Challenges

• CPGs and workflow
• inter- and intra organisational careflow support
• communication problems within and between HCO,
  produce
• distributed parallelised CfMS
• multi-agent research - overcoming limitations to
  traditional
• standard WfMS by introducing e.g. parallel
  activities, and complex, explicitly represented,
  negotiation processes among users

26
Conclusion Research Challenges

• CPG effectiveness, costs and socio-organisationa
  l considerations
• incomplete scientific evidence about benefits
  and harm
• scarce accurate cost data for clinical
  conditions and services

27
Guideline Support Tools Current Research
Discussion
K Dube Computer Science Dublin Institute of
Technology