Postcoordinated SNOMED CT Expressions in electronic patient records

1
Post-coordinated SNOMED CT Expressions in
electronic patient records

• Edward Cheetham
• NHS Connecting for Health Terminology Service

March 29th 2006
2
Outline

• Part 1 Introduction and background
• Part 2 Data entry
• Part 3 Storage, communication display
• Break
• Retrieval analysis
• Part 4 Transformation
• Part 5 Equivalence detection
• Part 6 Predicate optimisation

3
Introduction and background
Outline

• Post-coordination
• Definitions
• Motivation for post-coordination
• Scope
• Areas of strength v. exhaustive solution
• Logical equivalence v. realistic retrieval
• Patterns of post-coordination
• Data reality
• Modelling redundancy
• Model modelling incompleteness
• Information model overlap

4
Data entry
Outline

• Data entry options
• Definitions and principles
• General constraints
• model-based recording
• Specific constraints
• circumstantial constraints
• cardinality management
• Close-to-user data entry
• Input assist options
• Text processing/qualifier resolution
• Anticipatory/structured entry

5
Storage, communication display
Outline

• Storage options
• Stored forms immediate alternative
• Principles
• Expression repository
• Communication principles
• Display options
• Structured display
• Serialised renderings

6
Retrieval and analysis
Outline

• Definitions
• Retrieval and equivalence
• Normal and canonical forms
• Predicates and candidates
• Normal form transformations
• Generation rules
• Transformation algorithm walk-through

7
Retrieval and analysis
Outline

• Equivalence detection
• Normal form variants
• Absolute equivalence v. subtype detection
• Logical equivalence v. realistic retrieval
• Selective retrieval examples
• Canonical transformation
• Subtype testing vs. supertype testing

8
Retrieval and analysis
Outline

• Predicate optimisation
• Error patterns
• Modelling/data issues
• Over-modelling
• Incomplete modelling
• Sufficient versus necessary definitions
• Foundation model/value set assertions
• Primitive and defined concepts
• Mismatch between close-to-user guidance and
  transformation rules

9
Part 1
10
Introduction and background
Overview

• Post-coordination
• Definitions
• Motivation for post-coordination
• Scope
• Areas of strength v. exhaustive solution
• Logical equivalence v. realistic retrieval
• Patterns of post-coordination
• Data reality
• Modelling redundancy
• Model modelling incompleteness
• Information model overlap

11
Introduction background

• Post-coordination Definitions
• concept a clinical idea
• Concept a clinical idea to which a unique SCT
  ConceptId has been assigned.
• Expression a collection of references to one or
  more Concepts used to express an instance of a
  clinical idea.
• Post-coordination representation of a concept in
  an Expression using a combination of two or more
  Concepts

12
Introduction background

• Motivation for post-coordination
• Desirable characteristic size
• Completeness of coverage by Concepts
• Completeness of coverage by potential to
  represent concepts (as post-coordinated
  Expressions)
• Editorial/design principles
• Elective economy of distributed data, e.g.
• Laterality
• Severity

13
Introduction background

• Scope - Logical equivalence v. realistic/compound/
  complex retrieval
• Emphasis of this discussion on retrieval
  strategies based on the detection of logical
  equivalence
• Goal of complete concept/Concept definition
• realistic/complex retrieval
• Multiple OR clauses (ANDs within predicate
  expressions)
• Practical solutions will be a combination

14
Introduction background

• Scope - Areas of strength v. exhaustive solution
• Real and ideal data
• Ideal data would have the following properties
• Completeness and accuracy of modelling
• Non-redundancy of modelling
• Model-based solutions for all potential
  equivalence
• Known areas of reality/difficulty
• Overlap with information model solutions
• Prompt value solutions
• Model-independent equivalence (e.g. statuses)

15
Examples Post-coordination
Introduction background

• Post-coordination data-centric principles
• By qualification
• By sub-type refinement
• By concept combination

16
Examples Post-coordination
Introduction background

• Post-coordination
• By qualification
• Emergency appendectomy
• 80146002appendectomy
• 260870009priority25876001emergency
• By sub-type refinement
• Fracture of shaft of femur
• 71620000fracture of femur
• 363698007finding site71341001bone structure
  of femur

• 363698007finding site41111004bone structure
  of shaft of femur

17
Examples Post-coordination
Introduction background

• Post-coordination
• By concept combination
• laparoscopic cholecystectomy
• 38102005cholecystectomy
• 73632009laparoscopic procedure

18
Examples Post-coordination
Introduction background

• Post-coordination
• By concept combination Not valid
• vibration white finger
• 33679000vibration
• 371251000white
• 7569003finger

19
Examples Real and ideal data
Introduction background

• Ideal data
• 275504005 Lung cyst (disorder)
• Associated morphology Cyst
• Finding site Lung structure
• Fully defined (IsPrimitive False)
• Proximal Primitive Supertype Disease

20
Examples Real and ideal data
Introduction background

• Real data completeness and accuracy
• 275505006 Multiple lung cysts (disorder)
• Associated morphology Cyst
• Finding site Lung structure
• Primitive
• Associated morphology could be Multiple cysts
• Concept could be fully defined

21
InterludePrimitive and defined concepts

• Primitive A category of an ontology that cannot
  be (uniquely) defined in terms of other
  categories in the same ontology
• Necessary component of an ontology/terminology to
  make progress in development
• Introduce (some) properties/sanctions that cannot
  be represented in the SCT model
• Identification of the proximal (least
  semantically remote) primitive supertype(s) is
  consistent with the intention of introducing
  those supertypes in SCT
• However

22
InterludePrimitive and defined concepts

• not all primitives are the same
• Introduced primitives (natural kinds)
• procedure, disease
• most supporting hierarchies
• Unavoidable logical primitives (temporary
  natural kinds?)
• 177302005 wide excision of skin lesion
• Incomplete/arbitrary primitives
• 73759005 duodenectomy
• 85651007 plantar dissection
• but all are treated the same during the
  classification process and during the generation
  of canonical forms.

23
Examples Real and ideal data
Introduction background

• Real data redundancy of modelling
• Concept redundancy
• Modelling redundancy - wilful
• 75857000 Fracture of radius AND ulna (disorder)
• RG1 Associated morphology fracture
• RG1 Finding site bone structure of ulna
• RG2 Associated morphology fracture
• RG2 Finding site bone structure of radius

24
Relationship groups
Introduction background

• Add clarity to concept definitions
• Combines an attributevalue pair with one or more
  other attributevalue pairs
• For procedures, associate the correct method with
  the correct site

25
Examples Real and ideal data
Introduction background

• Real data redundancy of modelling
• Modelling redundancy - complex
• 95718009 rupture of lens (disorder)
• RG1 associated morphology rupture
• RG1 finding site lens clear
• RG2 associated morphology traumatic
  abnormality
• RG2 finding site lens clear

26
Examples Real and ideal data
Introduction background

• Real data information model overlap
• 191161006 Anemia due to ascorbic acid deficiency
• IsA Anemia
• Due to Ascorbic acid deficiency
• Due to association may be represented as
  related record statements using alternative
  record/information model mechanisms

27
Examples Real and ideal data
Introduction background

• Real data information model overlap
• 231968003 cataract following rupture of capsule
• IsA Cataract
• After Rupture of lens
• After association may be represented as related
  record statements using alternative
  record/information model mechanisms

28
Examples Real and ideal data
Introduction background

• Real data prompt value solutions
• 282462008 Ability to run 371150009 Able
• Can be related to
• 282474007 Able to run
• Interprets Ability to run
• Has interpretation Able

29
Examples Real and ideal data
Introduction background

• Real data prompt value solutions
• 282462008 First language 297499001 French
  language
• Should relate to
• 315571004 Main spoken language French
• Interprets 224076006
• No has interpretation attribute
• Demonstrable equivalence is probably the
  exception at present

30
Examples Real and ideal data
Introduction background

• Real data finding and activity statuses
• Not for resuscitation (finding)
• Cardiopulmonary resusn, not to be done
• Context-dependent procedure
• Associated procedure cardiopulmonary resusn
• Procedure context Not to be done
• No possibility of automatic equivalence
  detection
• Need for editorial stability for patterns such as
  these

31
Introduction and background
Recap

• Post-coordination
• Definitions
• Motivation for post-coordination
• Scope
• Areas of strength v. exhaustive solution
• Logical equivalence v. realistic retrieval
• Patterns of post-coordination
• Data reality
• Modelling redundancy
• Model modelling incompleteness
• Information model overlap

32
Part 2
33
Data entry
Overview

• Data entry options
• Definitions and principles
• General constraints
• model-based recording
• Specific constraints
• circumstantial constraints
• cardinality management
• Close-to-user data entry
• Input assist options
• Text processing/qualifier resolution
• Anticipatory/structured entry

34
Data entry

• Data entry - Definitions and principles
• (concept) model the editorially agreed set of
  defining and/or qualifying characteristics that
  can be used in formal concept definitions
• Allowable attribute attributes that can be
  applied to each main hierarchy or concept domain
  (according to the editorially-agreed model)
• Sanctioned attribute attributes that can be
  applied to each main hierarchy or concept domain,
  that have actually been modelled (as the
  attributes of defining or qualifying
  characteristics) in those domains.
• Unapproved attribute attributes that are
  currently not part of the concept model.

35
Data entry

• Data entry - General constraints
• Post-coordination with adherence to a model of
  sanctioned and/or allowable attributes
• Augmentation of such rules with, e.g.
  close-to-user recording strategies
• Manage complexity of distributed data
• Cleanse/simplify immediate record entry
• Maximise coded capture
• Possible error patterns
• Exceptions
• Optional representations

36
General constraints
Data entry

• Adherence to a model of allowable attributes
• TRG Appendix J allowable attributes

37
General constraints
Data entry

• Adherence to a model of allowable values
• TRG Appendix J value ranges

38
SCT Concept model
39
Recording options General constraints
Data entry

• Adherence to a model of sanctioned attributes
  and values
• Relationships table

40
General constraints

• Error patterns exceptions
• Sanctioned inevitable incompleteness
• Allowable
• Attributes specifying sites/lateralities etc.
  for observable entity concepts
• 363983007 Visual acuity
• 386708005 Visual acuity - left eye
• 386709002 Visual acuity - right eye

41
Recording options General constraints

• Error patterns exceptions
• Allowable
• Values
• Subject relationship context
• Person
• Organism
• Physical object

42
General constraints
Data entry

• Error patterns options/optionality
• How to represent route of administration?
• Extend current Method
• intravesical drug therapy?
• Extend current Approach
• via nasogastric tube?
• Introduce new Route of administration
• via antecubital vein?

43
Specific constraints
Data entry

• Post-coordination in particular settings
• Restrict recording choices
• Use-case dependent
• Can be revised over time
• Identify and remove options
• ? Less scope for revision
• Manage cardinalities
• Pre-adoption of unapproved attributes
• With care and planning
• With Editorial collaboration

44
Specific constraints
Data entry

• Post-coordination in particular settings
• By example
• Drug allergy decision support triggered by SNOMED
  CT substances
• Recording solution could based on
• Selected descendants of 416098002 Drug allergy
  (via causative agent xxx in distributed
  relationships)
• 416098002 Drug allergy with
• Causative agent xxx by post-coordination

45
Specific constraints
Data entry

• Post-coordination in particular settings
• Advantages
• May be beneficial to decouple record use from
  reference data
• Allows reference data refinement, with subsequent
  extension of recording solution to include wider
  set of concepts
• Disadvantages
• Without additional checks/steps, may overlook
  legacy records

46
Specific constraints
Data entry

• Identification and removal of options
• Route of administration example
• Unilateral decision to extend use of approach
  attribute (with, e.g. Extra-relationships)
• Advantages
• Sanctioned/Allowable
• Disadvantages
• May become incompatible with future model-based
  solution

47
Specific constraints
Data entry

• Pre-adoption of unapproved attributes
• Route of administration example again
• Adoption of route of administration routes of
  administration solution
• Work with CMDT/WG and Editorial Board
• Opportunity to
• discuss options and suitability
• Influence concept model development
• Disadvantage timescales, change of plan
• theres always pre-coordination

48
Specific constraints
Data entry

• Manage cardinality
• Is relationship cardinality specified by the
  number of relationship rows?
• How big can an Expression be?
• 297131007 Fracture of middle phalanx of finger
• Finding site
• Shaft of middle phalanx of middle finger
• Shaft of middle phalanx of ring finger

right
49
Specific constraints
Data entry

• Manage cardinality
• But
• 297131007 Fracture of middle phalanx of finger
• Finding site
• Shaft of middle phalanx of middle finger
• Shaft of middle phalanx of middle finger
• Valid laterality checks may be challenged
• Should these be separate concepts?

- right
- left
50
Close-to-user recording
Data entry

• Combination of sanctioned, approved (and variants
  of unapproved) recording to support
• Minimising complexity of and obstacles to
  recording
• Maximising extent and clinical faithfulness of
  SNOMED CT-encoded recording
• In a standard (predictable) form
• Distinct from, but transformable to, comparable
  normal forms

51
Close-to-user recording
Data entry

• Real data, certain patterns of ideal data
• Strict adherence to logical post-coordination
• Verbose immediate entry
• Un-selected
• ? Erroneous
• Complex default recording interactions
• Non-arbitrary mechanism for rationalising
  recording
• Do simple things simply, complex as required
• Supplementary steps for canonical transformations

52
Close-to-user recording
Data entry

• Create rupture of cortex of lens
• Refinement of site attribute/value for 95718009
  rupture of lens to structure of cortex of lens
• Close-to-user
• 95718009rupture of lens363698007finding
  site33347001structure of cortex of lens

53
Data entry

• Non-Close-to-User
• 95718009rupture of lens
• 116676008associated morphology125671007ruptu
  re,
• 363698007finding site33347001structure of
  cortex of lens
• 116676008associated morphology19130008trauma
  tic abnormality,
• 363698007finding site33347001structure of
  cortex of lens

54
Data entry

• Create right-sided fractured radius and ulna
• Addition of laterality right to 75857000
  Fracture of radius and ulna
• Close-to-user
• 75857000fracture of radius AND ulna
• 272741003laterality24028007right

55
Data entry

• Non-Close-to-User
• 75857000fracture of radius AND ulna
• 116676008associated morphology72704001fractu
  re,
• (363698007finding site23416004bone structure
  of ulna272741003laterality24028007right)
• 116676008associated morphology72704001fractu
  re,
• (363698007finding site62413002bone structure
  of radius272741003laterality24028007right)
  

56
Qualifier resolution

• One of several search strategy optimisations
• Word equivalents
• Duplicate term restrictions
• Qualifier resolution
• Pre-coordinated searches may not return a match
• Removal of common qualifier fragments from text
  search string to return focus concept
• Subsequent expression qualification
• Probably most predictable with limited value set
  axes chronicity, severity, urgency
• Rules/agreement needed for activation

57
Qualifier resolutionacute/chronic

• Acute reflux oesophagitis
• No matches returned
• Reflux oesophagitis
• Match 266433003 Reflux oesophagitis
• Reintroduce CourseAcute
• 266433003 26090800253737009

58
Qualifier resolutionemergency/elective

• emergency cholecystectomy
• No matches returned
• Cholecystectomy
• Match 38102005 Cholecystectomy
• Reintroduce PriorityEmergency
• 38102005 26087000925876001

59
Graphical/structured skull fracture

• Default focus
• 71642004
• Fracture of skull

Pathologic
71642004
60
Graphical/structured skull fracture

• Refine
• Finding site
• Bone structure of cranium
• To
• Parietal bone structure
• Or substitute
• 83385002
• Fracture of parietal bone?

Add Laterality Right
Add Morphology Fracture, stellate Or Wait
until compound status established (value
sets denormalised)?
Pathologic
71642004
83385002
83385002 27274100324028007
8338500227274100324028007, 11667600884359009
61
Data entry
Recap

• Data entry options
• Definitions and principles
• General constraints
• model-based recording
• Specific constraints
• circumstantial constraints
• cardinality management
• Close-to-user data entry
• Input assist options
• Text processing/qualifier resolution
• Anticipatory/structured entry

62
Part 3
63
Storage, communication display
Overview

• Storage options
• Stored forms immediate alternative
• Principles
• Expression repository
• Communication principles
• Display options
• Structured display
• Serialised renderings

64
Recording and storage

• Storage options - Stored forms
• Immediate
• Faithful and accurate
• What the clinician selected
• Immutable
• Alternative representations
• Can be refreshed with each release
• Can be generated according to evolving rules
• More in normal canonical forms

65
Recording and storage

• Storage options - Principles
• Schematic no right way, but
• Solutions specific to circumstances
• Use cases for data
• Features of data
• Stability of data e.g. nesting depth
• Options include
• Unrestricted relational representation
• Parseable text representation (compositional
  grammar)
• XML/object representations
• Restricted relational representation

66
Recording and storage

• Storage options - Principles
• Architectural no right way, but
• Storage of each Expression for each statement
• versus
• Storage of each Expression once (centrally),
  with multiple references from each statement
• Re-use (including e.g. renderings)
• Efficient re-computation of alternative forms
• Requires central solution

67
Storage Schematic options

• Unrestricted relational representation

• RGs, concept combination, depth

68
Storage Schematic options

• Restricted relational representation

69
Storage Schematic options

• Parseable text/object representation (SNOMED CT
  compositional grammar)
• 309431009arthroscopy of knee
• 272741003laterality24028007right,
• 260870009priority103390000elective

70
Storage Schematic options

• XML/object representations
• ltcode codeSystem"2.16.840.1.113883.6.96
  code"309431009" displayName"arthroscopy of
  knee"gt
• ltqualifiergt
• ltname code"272741003" displayName"laterality
  "/gt
• ltvalue code"24028007" displayName"right"/gt
• lt/qualifiergt
• ltqualifiergt
• ltname code"260870009" displayName"priority"/
  gt
• ltvalue code"103390000" displayName"elective"
  /gt
• lt/qualifiergt
• lt/codegt

71
Storage Schematic options

• Feature comparison
• Flexibility
• Nesting depth
• Attribute stability
• Retrieval performance
• Decision support
• Epidemiology

72
Storage Architectural options

• One statement One Expression
• Immediate/close-to-user entry
• Patient A, March 30th 2006
• 75857000fracture of radius AND ulna
• 272741003laterality24028007right
• Patient B, March 31st 2006
• 75857000fracture of radius AND ulna
• 272741003laterality24028007right

73
Storage Architectural options

• Alternative form
• e.g. a long normal transformation
• Patient A, based on January 2006 data
• 243796009context-dependent categories
• 246090004associated finding
• (64572001disease116676008associated
• morphology72704001fracture,
• 363698007finding site(23416004bone structure
  of ulna
• 272741003laterality24028007right)
• 116676008associated morphology72704001fractu
  re,
• 363698007finding site(62413002bone structure
  of radius272741003laterality24028007right)
  ),
• 408729009finding context410515003known
  present,
• 408731000temporal context410512000current or
  specified,408732007subject relationship
  context410604004subject of record

74
Storage Architectural options

• Alternative form
• e.g. a long normal transformation
• Patient B, based on January 2006 data
• 243796009context-dependent categories
• 246090004associated finding
• (64572001disease116676008associated
• morphology72704001fracture,
• 363698007finding site(23416004bone structure
  of ulna
• 272741003laterality24028007right)
• 116676008associated morphology72704001fractu
  re,
• 363698007finding site(62413002bone structure
  of radius272741003laterality24028007right)
  ),
• 408729009finding context410515003known
  present,
• 408731000temporal context410512000current or
  specified,408732007subject relationship
  context410604004subject of record

75
Storage Architectural options

• One record entry One Expression
• Advantages
• Tight coupling of record and original coded entry
• Disadvantages
• Redundancy of representation (esp. alts)
• Complex datatypes throughout record

76
Storage Architectural options

• Shared repository of Expressions, referenced or
  augmented by each statement
• Immediate/close-to-user entry
• Patient A, March 30th 2006
• 75857000fracture of radius AND ulna
• 272741003laterality24028007right
• Tested against shared store
• does not exist
• added to central store, given identifier
• cbfe5140-1bc5-11da-8cd6-0800200c9a66
• Identifier added to record

77
Storage Architectural options

• Immediate/close-to-user entry
• Patient B, March 31st 2006
• 75857000fracture of radius AND ulna
• 272741003laterality24028007right
• Tested against shared store
• Found
• cbfe5140-1bc5-11da-8cd6-0800200c9a66
• added to record

78
Storage Architectural options

• Shared repository could then contain
• ID cbfe5140-1bc5-11da-8cd6-0800200c9a66
• Expression 75857000fracture of radius AND
  ulna272741003laterality24028007right
• Alternative form e.g. long normal with most
  recent reference data
• 243796009context-dependent categories
• 246090004associated finding
• (64572001disease
• 408732007subject relationship
  context410604004subject of record

79
Storage Architectural options

• Classification of new Expressions
• Identification of predicate supertypes
• Concepts
• Other Expressions
• All, not proximal supertypes
• More in retrieval and analysis

80
Storage Architectural options

• Shared repository of Expressions with referencing
• Advantages
• Economic representation (esp. alts and classified
  data)
• Predictable (simple) datatype throughout record
• Disadvantages
• Possibility of separating record and code
• Combination

81
Communication guidance

• Essentially the same as for storage
• If the intention of the communication is the
  transfer of the original record entry, it is this
  form which should be communicated
• Stay as close to that entered and stored
• Explicitly grouped characteristics (if part of
  original record entry) should be communicated
• If there is a specific requirement to transfer a
  normal/alternative form, this may be communicated

82
Display rendering

• Any flexible recording solution creates its own
  problems
• Pre-coordination plus free text annotation
• Original text for parsed narrative
• Access to Fully-specified Names for synonym
  resolution/interpretation
• Post-coordinated expressions
• Re-presentation in unfamiliar surroundings
• Serialised rendering

83
Display rendering

• How do we get from here,
• (where a familiar structured
• input might be redisplayed)...

Pathologic
to here (where the same compound Concept would
be redisplayed serialised)?
84
Display rendering

• Need reproducible safe rendering rules
• Rudimentary rules proposed, but unvalidated
• Attribute names omitted
• Comma-separated
• Brackets around role groups
• Omit coarse-grained context-dependent focus terms

Previous example 8338500227274100324028007,
11667600884359009
85
Display rendering

• Need reproducible safe rendering rules
• Rudimentary rules proposed, but unvalidated
• Attribute names omitted
• Comma-separated
• Brackets around role groups
• Omit coarse-grained context-dependent focus terms

Previous example (all Preferred terms, comma
separated) Fracture of parietal bone,
LateralityRight, Associated morphologyFracture,
closed, stellate
86
Display rendering

• Need reproducible safe rendering rules
• Rudimentary rules proposed, but unvalidated
• Attribute names omitted
• Comma-separated
• Brackets around role groups
• Omit coarse-grained context-dependent focus terms

Previous example (all Preferred terms, comma
separated) Fracture of parietal bone, Right,
Fracture, closed, stellate
87
Display rendering

• Need generation/validation responsibility rules
• Workflow implications
• Natural human-readable not the only rendering
  requirement
• Exhaustive component visualisation
• Query predicate validation
• Nevertheless more to do

88
Storage, communication display
Recap

• Storage options
• Stored forms immediate alternative
• Principles
• Expression repository
• Communication principles
• Display options
• Structured display
• Serialised renderings

89
Part 4
90
Retrieval and analysis
Overview

• Definitions
• Retrieval and equivalence
• Normal and canonical forms
• Predicates and candidates
• Part 4 Normal form transformations
• Generation rules
• Transformation algorithm walk-through

91
Retrieval and analysis

• Definitions - Retrieval and equivalence
• (Selective) retrieval the detection in a
  clinical record or other resource of those
  Expressions that satisfy a set of query criteria
• Often includes suitable sub-type detection
• Equivalence the detection in a clinical record
  or other resource of those Expressions that
  exactly match a set of query criteria

92
Retrieval and analysis

• Definitions - Normal and canonical forms
• Normal form an alternative representation of a
  SNOMED CT Concept or Expression that makes easier
  logical comparison between concepts recorded
  using SNOMED CT.
• Simplified in terms of primitive Concepts
• Conjunctions and role restrictions
• Canonical form that normal form identified as
  being the standard alternative representation
  of a SNOMED CT Concept or Expression
• Includes specification for component ordering

93
Retrieval and analysis

• Definitions - Predicates and candidates
• Candidate Expression An Expression that is being
  tested to see if it is subsumed by another
  Expression.
• the data in the records
• Predicate Expression An Expression that is being
  tested to see if it subsumes another Expression.
• the predicate of the query

94
Retrieval and analysis

• Principles and practicalities - Retrieval v.
  Absolute equivalence detection
• Retrieval principle probably the more frequent
  approach for record interrogation
• select Expressions like
• Equivalence of use in
• Very precise Expression interrogation
• Content development (redundancy detection - as
  applied to Concept definitions)

95
Retrieval and analysis

• Principles and practicalities - Logical detection
  v. realistic retrieval
• Whether retrieval or equivalence, logical
  detection assumes ideal data
• One concept One Concept (if it exists), or
• One concept One canonical form
• Compound/complex retrieval is one strategy for
  managing real data
• OR statements in query specifications
• Requires knowledge/consideration of data
• May assume incomplete constraining of data entry
• Not mutually exclusive

96
Retrieval and analysis

• Principles and practicalities - Normal and
  canonical forms
• Generation rules general
• The application of a set of logical rules to an
  Expression, taking account of the definitions of
  the Concepts which it references
• Same principles apply to the generation of
  alternative views of Concepts in reference data
• Supertype alternatives
• Attribute alternatives
• Comprehensive v. minimal

97
Retrieval and analysis

• Principles and practicalities - Normal and
  canonical forms
• Generation rules previous
• (Generation of short canonical form)
• Identify proximal primitive supertypes
• Identify unshared defining characteristics
• Extension required for
• Incorporation of context, with inclusion of
  relevant information model variables
• Canonisation of values
• Manage arbitrary nesting depth
• Manage close-to-user Expression variants

98
Transformation steps

• Separate information model context

• Normalise Expression
• Separate focus Concepts from refinement
• Normalise attribute values in refinement
• Normalise focus Concepts
• Merge definitions merge refinement
• Create Expression

• Manage context
• Separate Expression context
• Resolve context
• Apply context

• (Additional steps for alternative forms)

99
Transformation steps

• Illustrative reference for a general Expression

100
Transformation steps

• Illustrative reference for a nested Expression

101
Transformation steps

• Comparison to normalised excision of
  infundibulum of right fallopian tube

71388002procedure 260686004method129304002
excision - action, 363704007procedure
site(8100009 structure of infundibulum of
fallopian tube 272741003laterality24028007 r
ight)
102
Transformation steps

• Illustrative reference for a context wrapper

103
Transformation steps

• Comparison to normalised no stiff neck

243796009Context-dependent categories 24609000
4associated finding 408729009finding
context 410516002known absent, 408732007subje
ct relationship context 410604004subject of
record, 408731000temporal context 410512000cu
rrent or specified
(161882006stiff neck 363698007finding
site 45048000neck structure)
104
Transformation steps

• Separate information model context

• Normalise Expression
• Separate focus Concepts from refinement
• Normalise attribute values in refinement
• Normalise focus Concepts
• Merge definitions merge refinement
• Create Expression

• Manage context
• Separate Expression context
• Resolve context
• Apply context

• Generation of alternative forms

105
Transformation steps
Separate information model context
Manage context
Normalize expression
Clinical statement or expression
Long normal form expression
Short normal form expression
Remove redundant attributes
106
Separate information model context

• Information that is not part of the SNOMED CT
  Expression may influence its interpretation.
• If none is present, the Expression is passed
  unchanged to the "normalise Expression" process.
• If the input is a structure that conveys
  additional contextual information (e.g. an HL7
  clinical statement)
• The Expression is separated and is passed to the
  "Normalise Expression" process.
• Relevant surrounding information model
  information is passed to the "Manage context"
  process.

107
Separate information model context
Some Act
.code Right knee arthroscopy
.mood RQO (request)
108
Separate information model context
Some Act
.code Normal range of knee movement
.mood GOL (goal)
109
Normalize expression

• Separate focus Concepts from refinement
• The set of focus Concepts in the Expression is
  passed to the "Normalise focus Concepts" process
• If the Expression contains a refinement, this is
  passed to the "Normalise attribute values in
  refinement" process

110
Normalize expression

• Normalise attribute values in refinement
• The value of every attribute specified is treated
  as an Expression and normalised
• Recursive normalization should be applied to all
  values even if they are represented by single
  Concepts.
• When all attribute values in the Expression
  refinement have been processed, the refinement is
  passed to the "Merge refinement" process

111
Normalize expression

• Normalise focus Concepts
• The set of focus Concepts is normalised to
  generated two separate outputs
• The set of normalised definitions of each focus
  Concept
• The non-redundant proximal primitive supertypes
  of the focus Concepts

112
Normalize expression

• Normalise Expression
• Merge definitions merge refinement
• The set of normalised definitions derived from
  the "Normalise focus Concepts" process are merged
  with one another to remove redundancy.
• The normalised refinement is merged with the
  pre-merged definition to create a single
  refinement
• Anticipated close-to-user consequences (such as
  indirect lateralisation and context-sub-typing)
  are resolved.

113
Normalize expression

• Normalise Expression
• Create Expression
• Combine the proximal primitive supertypes from
  "Normalise focus Concepts" process with the
  refinement derived from the "Merge refinement"
  process

114
Normalize expression Example patterns
115
Normalize expression Example patterns Key
Example Expression (compositional grammar)
Expression pattern illustrated
(Level x) recursion depth
Sub-processes
116
Normalize expression
71388002procedure AND 260686004method129304
002 excision ,363704007procedure
site66754008appendix structure
71388002procedure 260686004method129304002
excision ,363704007procedure
site66754008appendix structure
80146002Appendectomy
Single focus concept, no refinement (Level 0)
117
Normalize expression
260686004method129304002excision -
action ,363704007procedure site28231008gallbl
adder structure AND 260507000access129238008
endoscopic approach ,260686004method129433002
inspection - action ,363704007procedure
site83670000peritoneal cavity
structure ,370127007access instrument86174004
laparoscope
71388002procedure
260686004method129304002excision -
action ,363704007procedure site28231008gallbl
adder structure 260507000access129238008end
oscopic approach ,260686004method129433002ins
pection - action ,363704007procedure
site83670000peritoneal cavity
structure ,370127007access instrument86174004
laparoscope
71388002procedure AND 260686004method12930
4002excision - action ,363704007procedure
site28231008gallbladder structure 260507000
access129238008endoscopic approach ,260686004
method129433002inspection - action ,363704007
procedure site83670000peritoneal cavity
structure ,370127007access instrument86174004
laparoscope
71388002procedure 260686004method129304002
excision - action ,363704007procedure
site28231008gallbladder structure 260507000
access129238008endoscopic approach ,260686004
method129433002inspection - action ,363704007
procedure site83670000peritoneal cavity
structure ,370127007access instrument86174004
laparoscope
38102005Cholecystectomy 73632009Laparoscopic
procedure
Focus concept combination, no refinement (Level
0)
118
Normalize expression
120053002fallopian tube excision 363704007proc
edure site 2033006structure of right fallopian
tube
120053002fallopian tube excision 363704007proc
edure site 2033006structure of right fallopian
tube
31435000fallopian tube structure AND 272741003
laterality24028007right
31435000fallopian tube structure 272741003late
rality24028007right
260686004method129304002excision, 363704007
procedure site 31435000fallopian tube
structure AND 363704007procedure
site 31435000fallopian tube structure 2727410
03laterality24028007right
71388002procedure AND 260686004method129304
002excision, 363704007procedure
site (31435000fallopian tube
structure 272741003laterality24028007right)

71388002procedure 260686004method129304002
excision, 363704007procedure site (31435000fa
llopian tube structure 272741003laterality240
28007right)
Single focus concept, with unrefined, single
focus refinement (Level 0)
Single focus concept, with unrefined, single
focus refinement (Level 1)
Single focus concept, with unrefined, single
focus refinement (Level 0)

119
Normalize expression
120053002fallopian tube excision 363704007proc
edure site (8100009 structure of infundibulum
of fallopian tube 272741003laterality24028007
right)
120053002fallopian tube excision 363704007proc
edure site (8100009 structure of infundibulum
of fallopian tube 272741003laterality24028007
right)
120053002fallopian tube excision 363704007proc
edure site (8100009 structure of infundibulum
of fallopian tube 272741003laterality24028007
right)
120053002fallopian tube excision 363704007proc
edure site (8100009 structure of infundibulum
of fallopian tube 272741003laterality24028007
right)
260686004method129304002 excision -
action, 363704007procedure site31435000 fallo
pian tube structure MERGE WITH 363704007procedu
re site (8100009 structure of infundibulum of
fallopian tube 272741003laterality24028007ri
ght)
260686004method129304002 excision -
action, 363704007procedure site(8100009
structure of infundibulum of fallopian
tube272741003laterality24028007 right)
71388002procedure AND 260686004method129304
002 excision - action, 363704007procedure
site(8100009 structure of infundibulum of
fallopian tube272741003laterality24028007 ri
ght)
71388002procedure 260686004method129304002
excision - action, 363704007procedure
site(8100009 structure of infundibulum of
fallopian tube272741003laterality24028007 ri
ght)
Single focus concept, with refined, single focus
refinement (Level 0)
Single focus concept, with refined, single focus
refinement (Level 1)
Single focus concept, with refined, single focus
refinement (Level 2)
Single focus concept, with refined, single focus
refinement (Level 1)
Single focus concept, with refined, single focus
refinement (Level 0)

120
Manage context

• Separate Expression context
• If the normalised Expression contains context
  information, this is separated
• Resolve context
• Takes the information model context and the
  Expression context and attempts to resolve them
  to generate a single consistent context.
• Apply context
• Does or does not apply/replace context wrapper,
  depending on output of Resolve context process

121
Manage context
.mood RQO (request)
243796009context-dependent categories 36358900
2associated procedure( ),408730004procedu
re context 385644000requested 408731000tempor
al context 410512000current or
specified, 408732007subject relationship
context 410604004subject of record
71388002procedure 260686004method129304002
excision, 363704007procedure
site 66754008appendix structure
80146002Appendectomy
122
Manage context
.mood EVN (event)
243796009Context-dependent categories 24609000
4associated finding (161882006stiff
neck 363698007finding site 45048000neck
structure), 408729009finding context 410516002
known absent, 408731000temporal
context 410512000current or specified, 4087320
07subject relationship context 410604004subjec
t of record
243796009Context-dependent categories 24609000
4associated finding 408729009finding
context 410516002known absent, 408731000tempo
ral context 410512000current or
specified, 408732007subject relationship
context 410604004subject of record
(161882006stiff neck 363698007finding
site 45048000neck structure)
161881004No stiff neck
123
Transformation stepsSummary
124
Retrieval and analysis
Recap

• Definitions
• Retrieval and equivalence
• Normal and canonical forms
• Predicates and candidates
• Part 4 Normal form transformations
• Generation rules
• Transformation algorithm walk-through

125
Part 5
126
Retrieval and analysis
Overview

• Part 5 Equivalence detection
• Normal form variants
• Absolute equivalence v. subtype detection
• Logical equivalence v. realistic retrieval
• Selective retrieval examples
• Canonical transformation
• Subtype testing vs. supertype testing

127
Retrieval and analysis

• Principles and practicalities - Normal form
  optimisation
• Generation steps create long normal form
• Alternatives include
• Short normal form
• Long or short canonical forms
• Are there differences in relation to predicate
  a