Transcript Slide 1
Incorporating Data Mining
Applications into Clinical Guidelines
Reza Sherafat
Dr. Kamran Sartipi
Department of Computing and Software
McMaster University, Canada
{sherafr, sartipi}@mcmaster.ca
Computer-based Medical Systems (CBMS ’06)
June 22, 2006
1
Outline
Decision making based on data mining
results
Data and knowledge interoperability
Knowledge management framework
Tool implementation
Conclusion
Integrating data mining applications into clinical guidelines
2
Decision Making
Clinical
Practitioners
Decision
face
Support
criticalSystems
questions
(CDSS)
which
requires
decision
making:
– Computer
programs
– Provide
onlineofand
– The cause
a symptom
patient-specific
assistance
– Drug prescription
to health care professionals
– Treatment planning
to make better decisions
– Diagnosis of a disease
– … (many more)
– Clinical knowledge is stored
in a knowledge-base
Integrating data mining applications into clinical guidelines
3
Data Mining Applications
in Health care
Patient
Integrating data mining applications into clinical guidelines
4
Decision Logic
Condition
IF
the patient has had a heart stroke and is
above 50
THEN
Action
his health condition should be monitored!
Integrating data mining applications into clinical guidelines
5
Decision Logic (cont’d)
Decision making logic:
– Logical expressions
‘If-then-else’ structures
– Test for conditions
– Trigger actions
if ( (patient.age > 50) &&
(patient.previous_heart_stroke == true)
)
then …
Integrating data mining applications into clinical guidelines
6
Data Mining Decision Logic
Data mining
– Analysis and mining of data to extract hidden facts in
the data
– The extracted facts are represented in a data
structure called “data mining model”
Training vs. Application of a data mining model:
– Training the model: Building the model
– Application of the mode: interpreting for specific
patient data
Integrating data mining applications into clinical guidelines
7
Data Mining Decision Logic (cont’d)
Classification: mapping data into predefined classes.
(e.g., whether a patient has a specific disease or not)
Regression: mapping a data item to a real-valued
prediction variable. (e.g., planning treatments.)
Clustering: To identify clusters of data items. (e.g., to
cluster patients based on risk factors.)
Association Rule Mining: to find hidden associations in
the data set (e.g., how different patient data are related
based on shared relations such as: “specific diseases”,
“patients habits”, or “family disease history”.)
Integrating data mining applications into clinical guidelines
8
Data Mining Decision Logic (cont’d)
An example of regression model
[source:Otto,Pearlmen]
Vmax
3-4m/s
≥4m/s
≤ 3m/s
Doppler AVA
≤ 1 cm2
%100
%88
2-3+
%100
AVR recommended
1.1-1.6 cm2
AI severity
≥1.7 cm2
0-1+
%100
%100
%66
AVR not recommended
Integrating data mining applications into clinical guidelines
9
Application of Data Mining Results
Predictive Model Markup Language (PMML):
– XML based specification
– Meta model: Define the data structure of the model
– Different types of data mining models (clustering,
classifications, …)
– Extendable for model specific constructs
Share, access, exchange PMML documents
Integrating data mining applications into clinical guidelines
10
Proposed Health Care
Knowledge Management Framework
Phase 1: Build the data mining models
Knowledge Extraction
Guideline modeling
Guideline Execution
Integrating data mining applications into clinical guidelines
11
Proposed Health Care
Knowledge Management Framework
Phase 2: Encode data and knowledge
Knowledge Extraction
Data and knowledge
interoperability
Guideline Execution
Integrating data mining applications into clinical guidelines
12
Proposed Health Care
Knowledge Management Framework
Phase 3: Apply the knowledge for specific
patient data
Knowledge Extraction
Data and knowledge
interoperability
Knowledge Interpretation
Integrating data mining applications into clinical guidelines
13
Data and Knowledge Interoperability
HL-7 Reference Information Model (RIM)
Clinical Document Architecture (CDA)
– An XML-based standard for defining structured templates for
clinical documents
Data
– A general high level health care data model
Standard Terminology Systems (UMLS, SNOMED CT, etc)
Predictive Model Markup Language (PMML)
– An XML-based standard for representing data mining results
Guideline Interchange Format 3 (GLIF3)
– A clinical guideline definition standard
Integrating data mining applications into clinical guidelines
Knowledge
– Standard clinical vocabulary sets
14
Tool Implementation
A guideline execution engine based on GLIF
Logic modules apply data mining models and
are accessed through web services technology
Provides additional information to help guide the
flow in the guideline.
Integrating data mining applications into clinical guidelines
15
Conclusion
Data mining results can be used as a source of
knowledge to help clinical decision making.
We described an approach to apply different types of
data mining models in CDSS.
We used PMML and CDA for knowledge and data
representation.
A tool is developed that can interpret and apply the
mined knowledge.
We envision a future that data mining analysis results
are seamlessly deployed and used at usage sites.
Integrating data mining applications into clinical guidelines
16
Questions and Comments
Integrating data mining applications into clinical guidelines
18