Healthcare Data Interpretation Facility

1
Healthcare Data Interpretation Facility
Fabiane Bizinella Nardon University of São Paulo
Medical School Hospital São Paulo - SP -
Brazil fabiane.nardon_at_hcnet.usp.br
2
Beatriz de Faria Leão, MD, PhD1 Fabiane
Bizinella Nardon, MSc2 Miguel Artur Feldens,
MSc3 Altino Pavan, MSc4 Pablo Madril1
1Health Informatics Center - Federal University
of Sao Paulo 2University of Sao Paulo Medical
School Hospital
3Informatics Institute - Federal University of
Rio Grande do Sul, 4 Centro Educacional La
Salle de Ensino Superior,
Sao Paulo
Porto Alegre Canoas
3
Summary

• The HDIF model
• HDIF model versus HDIF RFP
• Issues to be discussed

4
HDIF model
Event Channel
HDIF
Observer
InputInformation
Methodology
Interpretation
EvokerObject
5
InputInformation
InputInformation
COASObservationDataObservation
6
Methodology
Methodology
Quantitative
Hybrid
Symbolic
7
Quantitative
Quantitative
BayesTheorem
DecisionTrees
ProbabilistNetwork
NeuralNetwork
CNM
Backpropagation
Art
8
Symbolic
Symbolic
KnowledgeRepresentation
InferenceMechanism
9
KnowledgeRepresentation
KnowledgeRepresentation
RuleSet
SemanticNetworkSet
FrameSet
CaseDataBase
1..
1..
1..
1..
Rule
SemanticNetwork
Frame
Case
10
InferenceMechanism
InferenceMechanism
ForwardChaining
BackwardChaining
DepthFirstSearch
(...)
11
Interpretation
Interpretation
SupplementaryInformation
KnowledgeRepresentation
ClinicalObservations
12
Evoker
EvokerObject
Activate the DSS when some event occurs
TimeBasedEvoker
MessageBasedEvoker
13
Observer
Observer
FuzzyModelObserver
DataSetObserver
DBSchemaObserver
RuleSetObserver
HistogramView
DataGrid
RuleHierarchyObserver
RuleListObserver
KnowledgeGraphObserver
14
Observer

• Represents the user interaction
• can be passive (just display the information)
• or active (gets information and sends a message
  through EventChannel)

15
HDIF model versus HDIF RFPMandatory Requirements

• The HDIF submission shall enable the HDIF to
  chain inputs and outputs to implement complex
  intelligent transforms.

Interpretation
InputInformation
KnowledgeRepresentation
ClinicalObservations
16
HDIF model versus HDIF RFPMandatory Requirements

• The HDIF submission shall provide interfaces
  enabling the HDIF to provide supplementary
  information supporting intelligent data
  interpretation. Examples of such supporting
  information include bibliographic references,
  automated references or reference documents and
  automated explanations.

Interpretation
SupplementaryInformation
17
HDIF model versus HDIF RFPMandatory Requirements

• For introspection beyond that supported by the
  OMG Interface Repository (IR) today, the HDIF
  submission shall specify HDIF-specific mechanisms
  for introspection. These mechanisms shall
  include interfaces that enable a HDIF component
  to reveal the following traits
• the type of intelligent data transform
• made available
• the technique, methodology, or algorithm
• implemented
• the level of utilization of the Lexicon Services

Yes, through Metodology Class
?
18
HDIF model versus HDIF RFPMandatory Requirements

• The HDIF submission shall be aligned with the
  CORBAmed work in progress focused on Clinical
  Observations Access Service (COAS).

InputInformation
Interpretation
ClinicalObservations
COASObservationDataObservation
19
HDIF model versus HDIF RFPMandatory Requirements

• The HDIF submission shall be aligned with the
  CORBAmed work in progress focused on the Lexicon
  Query Services (LQS).

COASObservationDataObservation
HDIF attribute Domain
LQS
KnowledgeRepresentation
20
HDIF model versus HDIF RFPOpcional Requirements

• HDIF submissions may utilize other types of data
  inputs in addition to the required inputs
  compliant with OMG CORBAmed COAS.
• Submissions may be extensible to accommodate
  intelligent transforms of other types of data in
  addition to clinical data.

InputInformation
COASObservationDataObservation
21
HDIF model versus HDIF RFPOpcional Requirements

• The HDIF submission may represent the confidence
  or uncertainty associated with output values.

Confidence
Interpretation
22
Issues to be discussed

• The HDIF submission may represent the confidence
  or uncertainty associated
• with INPUT values
• (meaning how much one truststhe information
  being entered)

23
Issues to be discussed

• An optional requirement
• In order to offer a performance measurement of
  the decision model the HDIF may depict the
  sensitivity, specificity, positive predictive and
  negative predictive values of the system as well
  as the costs and risks of its interpretation

24
Issues to be discussed

• Explanation
• What is automated explanation ?
• Explanation is dependent on the Knowledge
  Representation and on the Inference Mechanism,
  therefore maybe it should be aggregated to the
  Inference Mechanism ?

25
Issues to be discussed

• Should a HDIF depicts its domain ?

26
Issues to be discussed

• Arden Syntax ?
• It is not only a Knowledge Representation
  formalism
• Could a set of MLMs be considered as a HDIF ?

27
TEPR99 LATIN AMERICA
Rio de Janeiro, March 99