The SAGE Project Motivations, Approach, and Progress

1
The SAGE ProjectMotivations, Approach, and
Progress

• Samson Tu
• Stanford Medical InformaticsStanford University
  School of Medicine

July 2003
2
Outline

• SAGE Project Overview
• Scenario-Based Requirement Analysis
• Guideline Modeling
• Guideline Workbench
• Controlled Resources
• Deployment System
• Summary

3
SAGE Standards-Based Sharable Active Guideline
Environment

• 3-year NIST Advanced Technology Program grant
• IDX leads RD consortium that includes as
  partners
• Apelon, Inc.
• Stanford Medical Informatics (SMI)
• Intermountain Healthcare (IHC)
• University of Nebraska Medical Center (UNMC)
• Mayo Clinic
• Ultimate goal An infrastructure that will allow
  execution of standards-based clinical practice
  guidelines across heterogeneous clinical
  information systems (CIS)
• Focus is on the goal of deployment of guideline
  knowledge within the workflow of clinical
  information systems

4
SAGE Cast of Characters

• Apelon Eric Mays, Ron Nath, Tony Weida, Derrick
  Butler
• IDX Robert Abarbanel, Anne Saulovich, Graham
  Hughes, Guy Mansfield, Jeff Beers, John Foy,
  Julie Boyer, Larry Paulson, Malcolm Gleser,
  Mark Gaponoff, Nick Beard, Prabhu Ram, Rodney
  Wright, David Berg, John Thrun, Qin Ye
• IHC Roberto Rocha, Stan Huff, Rob McCluer, Craig
  Parker
• Mayo Pat Cahill, Chris Chute, Dave Mohr, Ruth A.
  Herman, George Klee, Kevin McDonald, Mark
  Nyman, Sidna Scheitel, Harold Solbrig
• SMI Mark Musen, Ravi Shankar, Samson Tu
• UNMC Jim Campbell, Karen Hrabak, Jim McClay, John
  Windle

Much of the work described in this talk are
contributions from other SAGE project members
5
SAGE The Problem to Be Solved

• Good News
• Clinical practice guidelines, based on
  evidence-based understanding of clinical best
  practices, are being produced at accelerating
  rate
• Bad News
• This pace far exceeds our ability to disseminate
  that knowledge, and far outpaces the ability of
  individual clinicians to keep up.

over use
practice variation
6
SAGE Solution Vision
Scenario-based workflow analysis informs
Creation of workbench to facilitate encoding and
localization
Encoding of guidelines using standard-based
guideline model
Development of generalizable deployment
architecture
Integration of guideline DSS and CIS using
standard controlled resources
7
SAGE Project End Point

• End-to-end demonstration of guideline encoding,
  localization, and execution at Mayo, IHC, and
  Nebraska

8
SAGE Contrast with Other Projects

• PRODIGY 3
• Multi-vendor GP guideline DSS deployed with
  single usage scenario
• EON/ATHENA
• Single-institution general guideline DSS deployed
  with single usage scenario
• InterMed
• General guideline DSS
• SAGE
• Multi-institution general guideline DSS
  demonstrated with multiple usage scenarios

9
SAGE Overall Methodology

• Focus on deployment-based requirements
• what will it take to implement encoded guideline
  knowledge within the workflow of clinical
  information systems
• Create useful deliverables
• Collaborate closely with leading Standard
  Development Organizations (e.g., HL7)
• Organize problem-focused work teams

10
SAGE Project Organization
(timeline approximate)

• Over-arching requirements
• Guideline Interactions
• CIS Actions

Iterative Guideline Model Specification
Guideline Model Requirements
End-to-end SAGE Integrated Testing
Iterative Workbench Design and Development
Guideline Workbench Requirements
Controlled Resources Requirements
Iterative Controlled Resources Development
Iterative Deployment Architecture Development
Deployment System Architecture Design
Deployment System Requirements
11
Outline

• SAGE Project Overview
• Scenario-Based Requirement Analysis
• Guideline Modeling
• Guideline Workbench
• Controlled Resources
• Deployment System
• Summary

12
Scenario-Based Requirement Analysis

• Goal Develop overarching requirements
• requirements common to all tiers of the
  guideline infrastructure, and which inform the
  requirements for each specific tier, including
  model, workbench, controlled resources and
  deployment system.
• Approach and protocol Structured walk-throughs
• Simulated scenarios of guideline usage in
  clinical care
• Identify
• Clinical information system (CIS)/guideline
  interactions
• CIS resources and actions
• Guideline decision-support services
• Workbench functions

13
Defining Workflow Requirements

• Assumption Guideline DSS is reactive
• Not in control of clinical workflow
• Respond to external events (including passage of
  time)
• Goals
• Empirically define points in care processes where
  guideline DSS may provide services
• Discover characteristics of human-computer
  interactions that enhances prospect of acceptance
• Approach and protocol
• Structured walk-throughs
• usability lab validation

14
Studying Workflow Work Cycle

• Walkthroughs
• Scenarios
• Develop CIS Screens
• Test CIS Screens in U-lab
• Modify Screens
• Use Cases
• UML Modeling
• Specification Document updates

This and the following 8 slides came from Dr.
Sidna Seitel of Mayo Clinic
15
1. Walkthrough

• Completed with Clinician input. Presented as
  PowerPoint slides

16
2. Scenario

• Developed for use in the u-lab. Input from
  physicians and project team

17
3. Develop CIS Screens

• Screens developed using VISIO/HTML

18
4. Usability Lab

• Done using scenarios and prototypes
• Captured on CD

19
5. Modify CIS Screens

• Screens modified based on results of the u-lab

20
6. Use Case

• Done using output from u-labs

21
7. UML Modeling

• Sequence diagram and use cases diagrams

22
8. Specification Document
23
Impact of Workflow Analysis (1) Organization
Model

• Inform necessary distinctions in organizational
  model
• Clinical settings, role, resources, event type
• Need to reconcile with standards (HL7)

24
Impact of Workflow Analysis (2)Model Guideline
Processes in Anticipation of Workflow
25
Specification of Context
26
Useful Deliverable

• A collection of decision-support use cases
• scenarios
• interactions
• decision-support opportunities
• Current focus
• Diabetes
• Community-acquired pneumonia
• Immunization
• Total-hip replacement
• Collaboration with HL7
• Commitment to donate use cases to Clinical
  Decision Support Technical Committee

27
Outline

• SAGE Project Overview
• Scenario-Based Requirement Analysis
• Guideline Modeling
• Guideline Workbench
• Controlled Resources
• Deployment System
• Summary

28
SAGE Guideline Modeling Overview

• Goal
• Requirement development
• Relationship to HL7
• Model of recommendation set

29
Create a Guideline Model that

• Is sufficient to encode guideline knowledge
  needed to provide situation-specific decision
  support and to maintain linked explanatory
  resource information for the end-user
• Uses standardized components that allow
  interoperability of guideline execution elements
  with the standard services provided within vendor
  clinical information systems.
• Includes organizational knowledge to capture
  workflow information and resources needed to
  provide decision-support in enterprise setting

30
Requirement Development

• Specify decision-support services through use
  cases
• decision-support services gt guideline model
  competency requirements
• Research state of guideline modeling
• Prior comparison papers
• Limited survey of use of workflow with guidelines
  and protocols
• Expose design decisions through prototyping

31
Adopt HL7 Guideline Standard Components

• Develop conceptual architecture of guideline
  components and locate areas of collaboration

Guideline Recommendation Set
Goals
Decisions
achieve/avoid
organizes
Context
Actions
Current HL7 CDS TC activities
Guideline Model
use
Expression Language
Guideline Concepts
Organization Model
32
Modeling Guideline Recommendation Set

• Original goal as HL7 activity To develop
  standardized flowchart model for human
  understanding and computer encoding
• Integration of decision making and activity
  specification
• Expressive process model allows sequencing,
  repetition, and concurrency (branching and
  synchronization) of decisions and activities
• Well-defined semantics

33
Proposal

• Flowcharts viewed as examples of organized sets
  of guideline recommendations
• Recommendation consists of
• Context (patient state, current therapy, clinic
  setting, provider)
• Decision (choice among alternatives)
• Action (order, referral, tests, procedure, )
• Two classes of recommendation set
• Decision map collection of decision points
• Activity graph network of activities organized
  over time
• Current work in HL7 Relate recommendation sets
  to HL7 Reference Information Model

34
Use of Workflow Process Model

• Workflow process model as basis of guideline
  process specification
• Workflow process model (WfMC, 1999)
• Network of activities and their relationships
• Formal semantics Mapping to Petri net
• However
• Not all guideline recommendations are processes

35
Decision Map Definition

• Collection of decision points
• Intent
• depict recommendations that are decision-centric
• Node types
• context
• decision
• action
• Properties
• no single entry node
• not necessarily connected
• not necessarily single-threaded
• Examples
• PRODIGY3 (SCHIN, UK)
• ATHENA (SMI/VA)

continue lifestyle change
hypertensive no medication
initiate med
substitute med
hypertensive w/ medication
increase dose
add drug
36
Decision Map Uses

• Designed for decision-support to single provider
  at single point in time
• determine relevant contexts and deal with
  decisions for those contexts
• other systems deal with workflow issues
• Specializations
• Condition/action statements (GEM)
• Augmented transition network (PRODIGY3)
• Connected graph
• One alternative allowed at decision point
• Decision tree
• An action is followed by a set of possible
  context nodes

37
Activity Graph Definition

• Network of activities organized over time
• Directed graph of
• nodes
• context
• decision
• action
• route
• transitions
• Specification of processes
• computational process
• care process

38
Activity Graph Properties

• Process
• unique entry node
• connected
• not necessarily single-threaded
• Action
• has states (enabled, active, completed)
• scheduling constraint
• completion condition
• may be repeated
• Transition constraints
• XOR/AND join
• XOR/AND split

Clt1.2
Get Creatinine
A
Cgt1.2
AND JOIN and XOR SPLIT
39
Sample activity graph from SAGE Project
40
(No Transcript)
41
Useful Deliverable A Guideline Model that is

• Built from ground up with available standards
• Recommendation sets with well-defined semantics
• GELLO expression language, HL7 data types,
  standard terminologies
• Workflow aware DSS will not be in control of
  workflow process
• Built with past modeling and deployment
  experiences in mind

42
Outline

• SAGE Project Overview
• Scenario-Based Requirement Analysis
• Guideline Modeling
• Guideline Workbench
• Controlled Resources
• Deployment System
• Summary

43
Develop a Workbench that

• Allows domain experts and knowledge engineers
  efficiently encode guideline logic, view, edit,
  manage, and test the library of computable
  clinical guidelines.
• Allows linking of encoded guidelines with
  explanatory resources
• Allows reuse of encoded guideline components
• Allows local stakeholders to import and adapt
  shared encoded guidelines and their associated
  knowledge bases for their local clinical
  information systems

44
Workbench Requirements

• Chose Protégé-2000 as guideline modeling platform
• baseline workbench automatically generated from
  guideline model
• Workbench requirement development
• asses features of Protégé-2000
• identify desirable features
• Evaluated 9 existing guideline workbenches
• based on published papers and available demo
  systems
• compared workbenches along a set of dimensions
• Analyzed life-cycle usage of guideline workbench

45
Guideline Knowledge-Acquisition Process
Guideline Model
Source Documents
Conceptualization
Use Case UML diagrams
46
Protégé-2000 As a Guideline KA Workbench

• Automatic generation of model-specific
  user-interface forms allows rapid prototyping
• No guidance for conceptualization and encoding of
  CPG in KBs
• Create wizards to support KA tasks
• Protégé-2000 is component-based and has an
  extensible architecture
• Access to external terminology through slot
  plug-in
• Validation through special tab plug-ins

47
KA Wizard (1) Explicit Modeling of Guideline KA
Process

• Scripts define sequence of KA tasks

48
KA Wizard (2) Creation of Alternative Views of
Protégé Forms

• Work around Protégés 1-class/1-form restriction

49

KA Wizard (3) Guiding a User through KA Tasks

• Based on specification of KA tasks and mapping of
  forms, the wizard presents a sequence of KA forms

• Top-down interview (TurboTax metaphor)
• Task-oriented recipe of how-to (Office
  paper-clip metaphor)

50
Accessing External Resources Terminology Plug-In

• A concept is represented as a terminological
  class
• A plug-in allows search and selection of terms
  from an external terminology server (developed by
  Apelon colleagues)
• Invoked everywhere that a controlled term is
  needed
• Cache references to controlled terms in Protégé
  to support browsing in absence of terminology
  server

51
Linking Guideline Recommendations with External
Documents

• An HTML widget to acquire URLs

52
Validation of Guideline KB

• Integrity constraints
• Local constraints
• constraints on slot values (e.g. type,
  cardinality)
• shown with red border in Protégé GUI
• Global constraints
• constraints that span across multiple slots,
  instances, or classes
• encoded in Protégés PAL constraint language
• Other types of validation (not done yet)
• Safety rules
• Conformance to guideline intentions
• Correctness of subsumption relationships

53
Local Constraints Validation Through
Facet-Constraint Tab

• Tab that allows checking of select classes for
  instances with facet constraint violations

Instance with facet constraint violations
Classes that has instances with facet constraint
violations
Summary of facet constraint violations
Classes to validatee
54
Global Constraints Validation Using PAL
Constraints
55
Indexing Retrieving Guideline Registry

• Technology based on ebXML
• Originally designed for e-business
• Function
• Submission, indexing, and retrieval of guidelines
• Version management (future attraction)
• Meta-data
• Dublin Core
• attributes from GEM

56
Problems in Use of Wizards

• Mapping and synchronization of guideline
  instances and wizard instances
• Wizard creates and display wizard-specific forms
  and instances for mapped instances in guideline
  KB
• Easy to partition a guideline instance into
  multiple wizard instances, difficult to aggregate
  multiple guideline instances
• Management and automation of domain-specific and
  book-keeping actions
• Specification of meaningful knowledge-acquisition
  subtasks

57
Problems in Use of Terminology

• Constraints on allowed terms
• Concepts used in guidelines are not always
  available as terms in existing terminology
• Primitive terms
• e.g. haemophilus influenza type b conjugate
  vaccine
• Compositional terms
• e.g. Progressive neurological finding isa
  Neurological finding Associated course
  Progressive
• e.g. Respiratory problems excluding asthma

58
Problems in Use of Constraints

• Insufficient facet constraints
• e.g. allowed values for Instance type
• Difficulty in extending PAL
• Not easy to add new predicates to PAL
• Weak constraint checking during editing process
• Facet constraints insufficient to specify allowed
  slot values
• PAL constraints not used for selecting slot
  values

59
Wish List

• Standardized terminology and information models
• Express constraints on legal codes
• Facilitate definition of new codes
• A plug-in architecture for developing and using
  wizards to perform specific tasks
• A plug-in architecture for defining and using
  alternative constraints in selecting and setting
  slot values
• A variety of constraint types
• A standardized way to invoke alternative
  constraint engines for checking legal slot values

60
Useful Deliverable

• Software tools for authoring, editing, encoding,
  and maintaining guidelines
• Support access to guideline content model,
  controlled resources, and related documents
• Provide aids for guideline encoding process
• Ensure complete encoding of guideline knowledge
• Support guideline registration, indexing, and
  retrieval

61
Outline

• SAGE Project Overview
• Scenario-Based Requirement Analysis
• Guideline Modeling
• Guideline Workbench
• Controlled Resources
• Deployment System
• Summary

62
Controlled Resources

• Purpose
• Standard data access and service layers have to
  be specified before complete interoperability
  between CCPGs and EMRs can be achieved
• Main focus
• Solve the Curly Braces Problem
• Virtual Medical Record (vMR)
• Clinical Expression Models (CEMs)
• Standard Coded Terminology
• Characterize Host System Services

63
Arden Syntax, the Curly Braces Problem

• data / creatinine in mg/dL / creatinine
  read last ('6210669','6210545','6000545','CREAT'
  
• logic
• __________________________________________________
  __
• If we had standard expression language, standard
  data model, standard terminology
• Example from GELLO (Boxwala, HL7 CDS TC
  Baltimore 2002)
• creatinine Observation.select(coded_concept
  "C0428279").last()
• Class Observation and attribute
  coded_concepts are parts of a patient data model

64
Standard Data Model Virtual Medical Record

• Patient data model that is simplification of
  medical record
• Only has distinctions important to DSS

65
Semantic Inter-operability

• Operations on VMR have to be interpretable to
  clinical information systems
• Approach Formulate operations on VMR as HL7
  version 3 message types

HL7 Domain Information Model of VMR
66
Current Work on VMR

• Prior work by Newcastle group
• Based on PRODIGY implementation
• Defines VMR in terms of a particular HL7 model
• SAGE work (led by IHC)
• Based on SAGE requirements
• Defines VMR as specializations of existing HL7
  models
• Reconciliation in progress

67
Standard Terminologies

• Emerging terminology standards
• SNOMED CT, LOINC,
• Compositional capability
• Emerging standard terminology services
• HL7 Common Terminology Service
• Open Terminology Service (Chute et al)

68
SAGE Terminology Requirements

• Guideline authoring-time requirements
• Selection of appropriate terms from standard
  terminology
• Composition of terms from standard terminology
• Extend terminology when appropriate concept not
  present
• Guideline execution-time requirement
• Resolve subsumption relationship

69
Selection of Terms at Guideline Authoring Time

• A term is represented as a HL7 concept descriptor
• An Apelon plug-in allows search and selection
  of terms from a terminology server

70
Composition of Terms from Standard Terminology

• ConceptExpression as Boolean combinations of
  existing concepts
• Boolean combinations as set operations within a
  terminology
• gtNo intractability issue with disjunction and
  negation

71
Extend Terminology When Appropriate Concept Not
Present

• Concepts used in guidelines are not always
  available as terms or Boolean combinations of
  existing terminology
• Primitive terms
• e.g. haemophilus influenza type b conjugate
  vaccine
• Compositional terms
• e.g. Progressive neurological finding isa
  Neurological finding 102957003 Associated course
  Progressive 255314001
• Need a process for recommending additions to
  terminologies
• Need precise description of terms

72
Clinical Expression Model

• VMR and terminology not sufficient to specify
  patient data
• Uniqueness
• Observation code ? immunization consent
  given,immunization consent refusedvalue ?
  present, absent
• Observation codeimmunization consent status
• value ? absent, consent given, consent
  refused
• Concept-specific constraints (e.g. units of
  measurements)
• Clinical Expression Model
• For a particular class of data, further constrain
  VMR
• Part of standardization process HL7 templates

73
Clinical Expression Model Example
Constraint on value of code slot
Constraint on value of allergen slot
74
Standard Data Access Summary

• Description of patient information is fully
  specified by
• VMR broad classes defines the structure of data
• CEM specialized constraints on data values
• Terminology particular concepts
• Formulation of data access as HL7 message types
• Requirement for data mapping from legacy
  institutional data repositories

75
Controlled Resource Deliverable (1)

• The number one barrier to the successful and
  widespread implementation of decision support
  tools is the Curly Braces problem
• Deliverable A recommendation of a standard
  model, terminology, and method for sharing data
  access specifications

76
Controlled Resource Deliverable (2)

• The second barrier is related to the complexity
  of the clinical data that is required for a CCPG
  to run
• Coded clinical data involves more than
  terminology codes
• Specifying aggregates such as blood pressure
  requires modeling
• Deliverable Development and implementation of
  conceptual models to represent the meaning of the
  clinical data, and a collection of services to
  store, retrieve, and analyze these complex
  language expressions

77
Controlled Resource Deliverable (3)

• The third barrier corresponds to the availability
  of specific services in the hosting clinical
  information system required by the decision
  support system
• Examples Retrieval of clinical data (vMR),
  message routing and notification (inbox, printer,
  pager), scheduling (appointments, visits,
  consults), placement of orders, etc.
• Deliverable Enumerate and formally define all
  the services required by a CCPG system, enabling
  its implementation as a separate but integrated
  component of a hosting clinical information system

78
Outline

• SAGE Project Overview
• Scenario-Based Requirement Analysis
• Guideline Modeling
• Guideline Workbench
• Controlled Resources
• Deployment System
• Summary

79
Big Challenge Guideline Deployment System

• Software that integrates electronic guidelines
  with the clinical information system to
  operationalize the guideline for clinicians

• Administer Download, import, store
• Localize Clinical edits, local constraints
• Binding Mapping to local terminologies,
  events, resources, roles, host system
  services
• Execute Activation of guideline via CIS
  workflow

80
Recap

• SAGE aims to create a infrastructure for
  implementation of guideline in workflow of
  clinical information systems
• Components
• Use cases of guideline-based decision support
• Workflow-aware and standard-based guideline model
• Guideline workbench with access to models and
  controlled resources
• Controlled resources that provides standard
  interface between guideline system with host
  information system
• Deployment architecture
• Close collaboration with HL7

81
SAGE A word from our sponsor . . .

• The National Institute of Standards and
  Technology (NIST), an arm of the U.S. Department
  of Commerce, funds high risk research through
  its Advanced Technology Program (ATP).
• The mission of the NIST/ATP program is To
  accelerate the development of innovative
  technologies for broad national benefit through
  partnerships with the private sector.
• NIST/ATP projects must entail research that
  leads to significant national benefits.

The SAGE project is partially funded by NIST/ATP
Cooperative Agreement Number 70NANB1H3049