modeling recommendation sets of clinical guidelines
DESCRIPTION
Modeling Recommendation Sets of Clinical Guidelines. Samson Tu Stanford Medical Informatics Stanford University School of Medicine HL7 Working Group Meeting Memphis, TN September 11, 2003. Outline. Original work-item goal Reformulated work item Review of past work - PowerPoint PPT PresentationTRANSCRIPT
Modeling Recommendation Sets of Clinical Guidelines
Samson TuStanford Medical Informatics
Stanford University School of Medicine
HL7 Working Group MeetingMemphis, TN
September 11, 2003
Outline Original work-item goal Reformulated work item Review of past work SAGE implementation experience Next steps?
Original work-item goal To develop standardized “flowchart”
model for human understanding and computer encoding Expressive process model
allows sequencing, repetition, and concurrency (branching and synchronization)
Integration of decision making and activity specification
Visual clarity Well-understood semantics
Background: problems What is a flowchart?
Different usages: not all diagrams represent processes
Workflow process model may not be appropriate for all
Study of guideline diagramming conventions
Tu, SW, Johnson, PD, Musen, MA (2002). A Typology for Modeling Processes in Clinical Guidelines and Protocols, Stanford Medical Informatics Report SMI-2002-0911
Reformulated work item “Flowcharts” viewed as organized sets of
guideline recommendations Recommendation consists of
context: clinical setting, patient state, current therapy, provider role, triggering event
decision: encapsulate decision-making knowledge action: computer or clinical actions
Two classes of “recommendation set” activity graph decision map
Common ground for various guideline models
Tu, SW, Campbell, J, Musen, MA, The Structure of Guideline Recommendations: A Synthesis, AMIA 2003 Symposium
Activity graph Used to specify
processes computational and care
processes branching and
synchronization (split/join) Directed graph of
context decision step action step route: purely for branching
and synchronization Adapted from Workflow
Management Coalition process model
unevaluatedCough (AND
split)
evaluate ACEas cause
treat presumedPNDS
order chestradiograph
coughgone?
abnormalCXR?
(AND join)
cough afterACE = yes
presumed PNDS = yes
(ANDsplit)
cough gone=no
coughgone?
Decision map Used to model
If/then statements Augmented transition network
Connected graph One alternative allowed at decision
point Decision tree
An action is followed by a set of possible context nodes
Collection of decision points context, decision, action nodes differ from activity graph in the
underlying computational model
hypertensiveno medication
initiatemed
continuelifestylechange
hypertensivew/ medication
substitutemed
adddrug
increasedose
?BP_controlled.eq(“true”)
Review of past work Proposal of using WfMC process model as
basis for “flowchart” (2001-05 Minneapolis) Visio stencil and template (2002-01 San Diego) Review of literature (2002-05 Baltimore) Mapping of guideline process to WfMC process
and then to Petri net (2002-05 Atlanta) Reformulation of work item as model of
“recommendation set” (2002-10 Baltimore) Partial reconciliation with HL7 RIM (2003-01
San Diego) Incorporation into SAGE guideline model
(2003-04 San Antonio)
Recommendation set in use: SAGE implementation experiment Selection of guideline Development of usage scenarios Specification of guideline logic, data
model, and terminology Encoding of guideline in Protégé-
2000 Simulation of implementation in IDX
clinical information system (CIS)
Selection of guideline and development of usage scenarios Immunization guideline selected for first
end-to-end experiment Four usage scenarios
Neonatal orders for immunization immediately after birth
Primary care immunization with standard care protocol in place
Primary care immunization, physician confirmation
Population-based reminders to patients or providers
Specification of guideline logic
Interpretation and operationization of guideline recommendations
Formulation of guideline logic sensitive to scenarios
guideline documents IF NO CONTRAINDICATION TO HEP B AND
NO REASON FOR DEFERRAL AND NUMBER OF HEP B VACCINE DOSES = 3 AND DOSE GIVEN WITHIN 7 DAYS OF BIRTH AND 3RD DOSE GIVEN BEFORE 6 MONTHS AGE AND TIME FROM LAST DOSE IS >= 8 WEEKS AND AGE < 19 YEARSTHEN ADVISE MONOVALENT HEP B VACCINE IM DUE
clinical scenarios
Specification of data model
A data model constrains how data is viewed and used in computable guideline
Detailed data models correspond to constraints on vMR classes
vMR classes
Concepts in guideline logic
Anaphylactic reaction to influenza vaccineis aAllergy where code is vaccines allergy allergen is influenza vaccine reaction is anaphylaxis
Specification of guideline terminology
Make reference to standard terminologies whenever possible
Need to define new terms
concepts in guideline logic
reference terminology
‘anaphylaxis’ = SNOMED CT:39579001contaminated wound lesion = ‘wound lesion’ with associated morphology ‘contaminated laceration’chronic pulmonary disease = ‘Chronic respiratory disease’ AND ‘Disease of lower respiratory system’
Encoding of guideline in Protégé-2000: top-level processes
Top-level activity graphs model DSS interactions with providers through clinical information system
Highly dependent on workflow
Immunization guideline logic as decision map
A decision involve choice of assertion to make about an immunization
Due and can be given Deferred Contraindicated or
not due Assertion about
immunizations tested at top-level activity graph
Lessons from SAGE experiment Top-level processes model guideline DSS
reactions/interactions with CIS guideline DSS not in control of workflow separation of workflow and medical knowledge
Decision map (coupled with a decision model) provides good cognitive correspondence with rule-like recommendations
Guideline encoding dependent on scenarios dependence on workflow dependence on formulation of recommendations
Limitations of SAGE experiment Multi-level multi-choice
recommendations not tested Refinement of medication recommendations Alternative classes of medications with
individual indications and contraindications Complex medical processes not tested
Concurrent processes that require branching and synchronization
Multi-encounter processes
Next step? Need for RIM harmonization? Specification for balloting? Date for closure?