Semanticsbased Knowledge Management for Translational Medicine

1
Semantics-based Knowledge Management for
Translational Medicine

• Tonya Hongsermeier, MD, MBA
• Corporate Manager, Clinical Knowledge Management
  and Decision Support
• Vipul Kashyap, PhD
• Senior Medical Informatician, Clinical Knowledge
  management and Decision Support
• Clinical Informatics RD
• Partners Healthcare System

2
The Volume and Velocity of Knowledge Processing
Required

• Medical literature doubling every 19 years
• Doubles every 22 months for AIDS care
• 2 Million facts needed to practice
• Genomics, Personalized Medicine will increase the
  problem exponentially
• Typical drug order today must account for Age,
  Weight, Height, Labs, Other Active Meds,
  Allergies, Diagnoses
• Today, there are 3000 molecular diagnostic tests
  on the market

Covell DG, Uman GC, Manning PR. Ann Intern Med.
1985 Oct103(4)596-9
3
Where are most healthcare systems today?

• Knowledge hardwired into applications, not
  easily updated or shared
• Most EMRs have a task-interfering approach to
  decision support, sub-optimal usability
• Knowledge-engineering tools typically designed to
  receive data, not support acquisition or
  maintenance
• Consequently, clinical systems implementations
  are under-resourced with adequate knowledge to
  meet current workflow and quality needs
• Lack of healthcare leadership or resource
  investment in processes for knowledge acquisition
  and management
• Doesnt bode well for personalized medicine

4
Knowledge Management and Decision Support
Intersection Points in Translational Medicine
Test ordering and documentation guidance
Patient Encounter (direct care or clinical trial)
Clinical Trials Referral
Personalized Medicine Decision Support Knowledge
Repository
Tissue-bank
Structured Test Result Interpretations
Therapeutic ordering and documentation guidance
Knowledge Discovery, Acquisition Management
Structured Research Annotations
Bench RD
Integrated Genotypic Phenotypic Database
Clinical Trials 1- 4
Pharmacovigilance
5
Example Hypertrophic Cardiomyopathy

• Autosomal Dominant inheritance
• Sudden Death risk (Reggie Lewis)
• gt200 Mutations involving 10 genes thus far
• Great phenotypic variability of expression
• Child and late-adult onset modes
• American College of Cardiology recommends
  clinical testing and follow-up of 1st and 2nd
  degree family members, new genetic tests now
  available

6
Dr. Genomus Meets Basketball Player Who Fainted
at Practice

• Clinical exam reveals abnormal heart sounds
• Family History Father with sudden death at 40,
• 2 younger brothers apparently normal
• Ultrasound ordered based on clinical exam reveals
  cardiomyopathy

Structured Physical Exam
Structured Family History
Structured Imaging Study Reports
7
Actionable Decision Support inthe Workflow
Context
8
Echo results triggers guidance to screen for
possible mutations

• beta-cardiac Myosin Heavy Chain (MYH7)
• cardiac Myosin-Binding Protein C (MYBPC3)
• cardiac Troponin T (TNNT2)
• cardiac Troponin I (TNNI3)
• alpha-Tropomyosin (TPM1)
• cardiac alpha-Actin (ACTC)
• cardiac Regulatory Myosin Light Chain (MYL2)
• cardiac Essential Myosin Light Chain (MYL3)

9
American College of Cardiology Guidelines for
Follow-up
Therapeutic Protocols
Predictive Monitoring Protocols
10
Knowledge Acquisition

• Connecting Dx, Rx, Outcomes and
• Prognosis Data to Genotypic Data for HCM

Gene expression in HCM Test Results
person
concept
date
raw value
Z5937X
3/4
Outcomes calculated every week
Syncope
microarray (encrypted)
Myectomy
ER visit
Z5937X
3/4
Atrial Arrhythymi
Palpitations
Z5937X
3/4
ER visits
Gene-Chips
Z5937X
3/4
Clinic visits
Ventricular Arrhy
Echocardio
Z5937X
4/6
ICD
Gene-Chips
Z5956X
5/2
Cong. Heart Failure
microarray (encrypted)
Cardiomyop
Z5956X
5/2
Atrial Fib.
Z5956X
5/2
Echocardio
Z5956X
5/2
EKG
Z5956X
3/9
Cardiac Arr
Z5956X
3/9
ER Visit
Z5956X
3/9
Thalamus
Z5956X
3/9
11
KM for Translational Medicine Functional/Busines
s Architecture
RD
CLINICAL TRIALS
DIAGNOSTIC Svs LABs
CLINICAL CARE
PORTALS
LIMS
EHR
APPLICATIONS
ASSAYS ANNOTATIONS
DIAGNOSTIC TEST RESULTS ASSAY INTERPRETATIONS
ORDERS AND OBSERVATIONS
KNOWLEDGE and WORKFLOW DELIVERY SERVICES FOR ALL
PORTAL ROLES
Genotypic
Phenotypic
DATA REPOSITORIES AND SERVICES
State Management Services
Semantic Inferencing And Agent-based Discovery
Services
KNOWLEDGE ACQUISITION AND DISCOVERY SERVICES
Knowledge Asset Management and Repository Services
Workflow Support Services
Data Analysts and Collaborative
Knowledge Engineers
Collaboration Support Services
Logic/Policy Domains
Meta- Knowledge
Knowledge Domains
Data Domains
NCI Metathesaurus, UMLS, SNOMED CT, DxPlain, ETC
other Knowledge Sources
Decision Support Services
INFERENCING AND VOCABULARY ENGINES
12
Key Functionalities/Technologies

• Data Integration
• Semantics and ontology based approaches
• Across multiple data sources
• Genotypic LIMS
• Phenotypic EHR
• Actionable Decision Support
• Semantic Inferences
• Management of Patient State
• Knowledge Acquisition, Maintenance and Evolution
• Ontology-based Definitions Management
• Ontology Change and Versioning
• Impact on Knowledge Assets

13
Decision Support Architecture
Maintenance Context
Transactional Context
Client
Editor
Event Broker
Content Management Server
OWL Ontology Engine
SWRL Rules Engine
Ontology Management Server
Rules Management Server
TRANSACTIONS
PUBLISHING
Knowledge Discovery Server
Knowledge Repository
Laboratory Information Mgt System (LIMS)
Electronic Health Record (EHR)
14
Clinical Knowledge
Genomic Knowledge
15
RDF/OWL Representation
has_name
has_problem
Harry
has_family_history
has_name
has_relative
sex
relationship
Father
onset_age
degree
has_family_history
type
first
is_living?
has_problem
has_family_history
.
16
SWRL Representation

• IF patient has family history of sudden death in
  a first degree relative
• before the age of 50
• THEN consider an echocardiogram for the patient
• patient(?p) has_family_history(?p, ?f)
  has_relative(?f, ?r)
• has_problem(?f, Sudden Death)
• degree(?r, first) onset_age(?r, ?age) ?age
  lt 50
• order_test(?p, ?t1) has_name(?t1,
  Echocardiogram)
• IF echocardiogram results are positive
• THEN order an HCM Molecular Diagnostic test for
  the patient
• patient(?p) has_structured_test_result(?p, ?t)
  
• test_name(?t, Echocardiogram) test_result(?t,
  positive)
• order_test(?p, ?t2) has_name(?t2, HCM
  Molecular Diagnostic Test)

17
State Management may also order test if Echo is
negative
Order Echocardiogram
Patient State
has_family_history(p1, f1) has_problem(f1,
Myocardial Infarction) has_relative(p1, f1,
r1) degree(r1, first) age_onset(r1,
45) relationship(r1, Father)
No
Positive?
Yes
Yes
No
Order HCM Diagnostic Test
Positive?
Add has_structured_test_result(p1, t1) name(t1,
Echocardiogram) test_result(t1, positive)
Yes
Add has_mol_diagnostic_test_result(p1,
t2) name(t2, HCM Diagnostic Test) test_result(t2
, positive)
Patient State and Potential Inferences Cached as
a State Network On the Client Side!
18
Content Discovery and Maintenance

• Collect longitudinal clinical observations
• Collect molecular diagnostic observations.
• Build models for predictive value of tests,
  interventions and outcomes
• For example, the model may predict that an
  certain missense is a better predictor of
• Late onset Dialative Cardiomyopathy (DCM), as
  opposed to
• Early onset Dialative Cardiomyopathy
• Reflected in updated rule

19
Knowledge Maintenance

• IF Molecular Diagnostic reveals MYH7 missense
  Ser532Pro or Phe764Leu
• THEN perform DCM monitoring and management
  protocol
• IF Molecular Diagnostic reveals MYH7 missense
  Ser532Pro
• THEN perform late onset of DCM monitoring
  protocol
• If MYH7 missense Phe764LEU
• THEN perform early onset of DCM monitoring
  protocol
• Rule Management Lifecycle
• Includes publishing of latest version of rules
• Rule Management Workflows
• Rule Versioning and Auditing

ROLLBACK
CHANGE
20
Knowledge Management Platform
Asset Management
Collaboration
2005
Knowledge Life-Cycle (CMS)
Ontology Management
2006
Meta-Knowledge Repositories
2004
Decision Support
Legacy Knowledge Editors
Legacy Application Knowledge Repositories
Discovery and Optimization
State Management Ontology Engine Inference
Engine Workflow Management
Business Intelligence/ Data Mining
Enterprise Clinical Services Knowledge Transaction
Repositories
Enterprise Clinical Services
2006
2005
21
Market Drivers will Continue to Make Knowledge
Management an Imperative for Translational
Medicine

• Genomics personalized medicine will require
  decision support architectures that can
  proactively support complex decision making
  answering 1,000,000 before run-time
• These systems will require self-adaptive, machine
  learning modes of knowledge acquisition, purely
  human dependent knowledge acquisition will not
  scale
• Knowledge-bases will need to support care-giver
  and consumer-based decision support requirements
• Pharma will need cooperative relationships with
  HIT vendors to make speed the translational
  medicine life-cycle

22
Acknowledgements

• The Construct tool from Cerebra was used to
  create OWL models
• Alfredo Morales helped create the OWL models