Building Blocks for Decision
Support in HL7
Samson W. Tu
Stanford Medical Informatics
Stanford University School of Medicine
Stanford, CA
Outline


HL7 Clinical Decision Support TC
Representation of DS knowledge




Arden Syntax
Clinical practice guideline (CPG)
Building blocks of CDS based on CPG
Messages
HL7 Clinical Decision Support
TC: History and Structure




Arden Syntax first met as part of HL7 in 1997
Version 2.0 formally transferred to and adopted
by HL7 and ANSI in August, 1999
CDS TC synonymous with Arden Syntax up to
2000 St Louis meeting
Current Structure



Arden Syntax SIG
Clinical Guideline SIG
Clinical Trial SIG
CDS TC Membership

Vendors


Healthcare Providers


Micromedex, Siemens, SMS, Eclipsys,
Epic...
LDS Hospital, Intermountain Health Care...
Academic

Newcastle, Harvard, Columbia, Stanford,
Yale, UT-Houston, ...
Shared Interests of CDS TC


Single-patient-focused healthcare
Common interest in decision-support
knowledge representation



Arden Syntax
Clinical practice guidelines as marked-up
documents and as computable knowledge base
Development of standards based on HL7
methodology and collaboration
Outline


HL7 Clinical Decision Support TC
Representation of DS knowledge




Arden Syntax
Clinical practice guidelines
Building Blocks of CPG
Messages
Arden Syntax and Medical
Logic Module


MLM designed to share medical knowledge
required for making a single decision
Mature standard




Decade-long evolution
Supported by multiple vendors
Continuously refined
Current projects



Fuzzy-logic extensions of Arden Syntax
XML format - multiple levels
Object-oriented data model and expression language
Challenges for Arden Syntax

Simple data model


Curly braces problem



Variables with time-stamped values
No structured way to access institutionspecific
MLM-specific mappings
Integration with HL7 version 3 RIM and
methodology
Representation of Clinical
Practice Guidelines for CDS


Current focus of decision-support research
Numerous formalisms for representing
guidelines


PROforma, PRODIGY, PRESTIGE, PacMan/Guide,
EON, Asbru, GLIF, …
CDS TC takes a component view


Has not endorsed a particular formalism
Work on building blocks
Building Blocks of DS Using
Clinical Practice Guidelines
Guideline
Goals
Processes
achieve/maintain
organize
Actions
Guideline Model
EMR/Host Systems
Decisions
uses
…
Expression
Language
depends on
use
Medical Concept
Model
Patient
use/build on
Data Model
Standard Medical
(VMR)
Terminology
CPG Component: Goals

Difficult to extract from existing guidelines



Often not explicitly stated
Need to be operationalized
Research issues


Representation
Multiple uses




For comparison with clinicians’ goals
As reason for making choice of intervention
As constraints
Patient-specific goals
CPG Component: Decision
Model


Making choice among alternatives
Many alternative models


Conditional rule
Pros-and-cons argumentation


With or without weighting
Decision-theoretic approach

Allows patient preference
CPG Component: Action Model


Specification of acts to be performed
Issues to resolve

System vs. medical actions


Temporal specification


e.g. prescribe an ACE inhibitor
Use of standard medical terminology


e.g. return appointment in 3 months
High-level action and refinement


e.g. send a message vs. order a test
e.g. order a specific laboratory test
Opportunity to leverage HL7 products (e.g.
Act class in RIM and work of O/O TC)
CPG Component: Process
Model

Organizes decisions and actions over
time




Sequencing, repetition and concurrency of
actions and decisions
Hierarchical decomposition of processes
Different kinds of processes
Represented as diagrammatical
flowchart
Example of Guideline
Management Process
Concurrent
actions
Action
step
If-then-else
decision
Alternative
choices
Patient
scenario
Subguideline
Toward a Common Process
Model

Requirements


Integration of decision making and activity
sequencing
Expressiveness




Allows sequencing, repetition, and concurrency
(branching and synchronization)
Visual Clarity
Well-understood and formal semantics
Problems with current guideline models



PRODIGY3: Augmented Transition Network
PROFORMA: Constraint Network
EON/GLIF: Operational semantics
Workflow as Basis for
Common Process Model

Workflow management



Workflow Process Definition Language



Controlling, monitoring, optimizing and supporting
business processes
Explicit representation of the business logic that
allows computer support
Version 1.1 (October 29, 1999)
Defined Workflow Management Coalition
Good properties


Formal semantics (mapping to Petri net)
Concordance with the RIM
Integration of Decisions and
Workflow Activities


Add clinical states and
decision steps as nodes
in flowcharts
Separate decisionmaking and workflow
functionalities
CPG Component: Expression
Language

Core of decision-support knowledge



If systolic BP > 140 and presence of DM and ACE
inhibitor is not contraindicated…
an episode of uncontrolled blood pressure that
follows the initiation of lisinopril within two
weeks…
Requirements




A data model for patient information
Reference to medical concepts and knowledge
Expression of temporal relationships
Use of abstraction
Old and New Expression
Languages

Arden Syntax as guideline expression
language



Mapping of complex datatypes into Arden
Loss of expressiveness
New object-oriented expression language
being developed



Use of HL7 datatypes
Use VMR
Built-in classes and methods and declaration of
new ones
CPG Component: Medical
Concept Model

Defines:



medical concepts used in guidelines
relationships among concepts
abstractions



Classifications (diabetes mellitus)
Deriving categorical values from numerical data
(abnormally elevated serum cholesterol)
Defining concepts in terms of others
Why Not Use Standard
Vocabulary Directly?


Need strict classification
hierarchy
Allows definition of
guideline-specific concepts
and axis of classification
DM not taking ARB:
Presence of DM and absence of
prescription for angiotensin II
receptor blockers

Organizes mappings to
host-system data
dictionary
Diabetes
DM type 1
DM not
taking ARB
Relationship to Standard
Vocabulary

Possible relationships



Vocabulary service as source of medical
concept model
Host system vocabulary as target of
terminology mapping
Challenges


Leverage work of done elsewhere
Distinguish vocabulary service from
functions of knowledge base
CPG Component: Patient Data
Model

Must be standardized



support expression language
support mapping to multiple EMRs
VMR talk at this meeting
Outline


HL7 Clinical Decision Support TC
Representation of DS knowledge




Arden Syntax
Clinical practice guidelines
Building Blocks of CPG
Messages
Messages Between DSS and
External Agents

Arden Syntax and MLM



Input: triggering event and clinical data
Output: alerts and reminders
CDSS for guideline-based care

Rich collection of messages needed


Recommendations for interventions, decision to
select a choice, requests for data, …
Much remain to be done
Summary



Arden Syntax and MLM a mature standard
with industry support
Current work centered around standardization
of format and infrastructure needed for
guideline-based case
Standardization of clinical guidelines still in
infancy


Current work looking at components of guidelines
Requires collaboration with many other
committees
Descargar

Document