Knowledge Representation in Practice: SNOMED CT - KR-MED.org

174
Knowledge Representation in Practice: SNOMED CT Kent A. Spackman, MD, PhD Chief Terminologist, IHTSDO Clinical Professor, Oregon Health & Science University, Portland Oregon USA KR-MED 2008 Tutorial Phoenix – May 31, 2008

Transcript of Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Page 1: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Knowledge Representation in Practice: SNOMED CT

Kent A. Spackman, MD, PhD

Chief Terminologist, IHTSDO

Clinical Professor, Oregon Health & Science University, Portland Oregon USA

KR-MED 2008 TutorialPhoenix – May 31, 2008

Page 2: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Tutorial 2nd Half Topics

• Description logic basics• Concept model domains, ranges and attributes• Application of additional description logic features to

special requirements: – Right identities: partonomies (SEP)– Role groups: complex procedures, disorders, situations– Role hierarchies: direct and indirect objects; causal and associational relationships

• Composition and syntax– normal forms– SNOMED compositional syntax, KRSS, and OWL

• Current redesign efforts – substances, observables, anatomy, events, conditions, organisms

• The context model & negation• The terminology / information model interface

2

Page 3: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

DESCRIPTION LOGIC BASICS

3

Page 4: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

What is description logic?

• Mathematical viewpoint:– A family of logics characterized by

• Formal set-theoretic semantics• Proofs of correctness and completeness of computation• Proofs of algorithmic complexity (PSpace, NP-complete,

NExpTime, etc)

• Knowledge representation viewpoint:– A set of constructs for representing terminological

knowledge (that which is always true of a meaning)– Algorithms and their implementations for performing:

• Subsumption (testing pairs of expressions to see whether one is a subtype of the other & vice versa)

• Classification (structuring a set of expressions according to their subsumption relationships)

4

Page 5: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Constructs for a very simple DL: EL

Name of construct Notation Semantics Primitive concept name A AI⊆∆I Primitive role name R RI⊆∆I×∆I Top (everything) > ∆I Conjunction (AND) CuD CI∩DI Exists restriction ∃R.C {x|∃y.RI(x,y)∧CI(y)}Concept definition (sufficient) A≡C AI=CI Primitive concept introduction AvC AI⊆CI

5

Page 6: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Reading DL expressions

u is the symbol for Boolean conjunction

– Short name: “and”– Operands are placed on either side: C u D– Both operands C and D must be true in order for the expression to be

true

∃ is the symbol for existential restriction

– Short name: “some”– Operands follow the backwards E symbol: ∃R·C

• R represents a “role” or relationship• C represents the value or target of the relationship

– Requires that there be an instance of relationship named R to someinstance of the class named C in order for the expression to be true

6

Page 7: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

The logic of green frogs

How could you use EL to define a green frog?

First, some concept names (A in the EL notation table):FrogGreenGreenFrog

Then a role name (R in the EL notation table):hasColor

Now introduce a definition of GreenFrog, using Frog, Green, and hasColor, along with C u D and ∃R·C as follows:

GreenFrog ≡ Frog u ∃ hasColor.Green

7

Page 8: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

The logic of green frogs

How do we read these expressions?

Usual DL syntaxGreenFrog ≡ Frog u ∃ hasColor.Green

KRSS syntax:(defconcept GreenFrog (and Frog (some hasColor Green)))

Every instance of the class named GreenFrogis an instance of the class named Frog which alsohas a “hasColor” relationship to an instance of the class named “Green”

And, every instance of the class named Frog which alsohas a “hasColor” relationship to an instance of the class named “Green”is an instance of the class named GreenFrog.

8

Page 9: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

The logic of green frogs

Right hand side only:

Usual DL syntaxFrog u ∃ hasColor.Green

KRSS syntax:(and Frog (some hasColor Green))

SNOMED compositional syntax:Frog: hasColor = Green

an instance of the class named Frog which alsohas a “hasColor” relationship to an instance of the class named “Green”

Exercise for the reader: how do you represent frogs that arecompletely green?

9

Page 10: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

A SNOMED example

• Headache is-a ache: finding-site = head structure – (and headache is marked as “defined” in concepts table).

• The class “headache” is sufficiently defined as the set of instances of the class “ache” which also have at least one finding-site relationship to an instance of the class “head structure”.

• And all instances of class “ache” with some finding-site relationship to an instance of “head structure” are instances of “headache”.

• Now, is that what you mean when you say “headache”?

10

Page 11: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

EXPRESSIONS & COMPOSITION

11

Page 12: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

SNOMED CT Expressions

• SNOMED CT coded information consists of structured (composed) collections of concept codes– These are called expressions– The meaning of an expression depends on the situation in which it is used

Example• The SNOMED CT code for “fracture of femur” represents the meaning

of “a break in a femur”– Depending on where it is used in a patient record, the code may mean

• The patient has a fractured femur• The patient’s main diagnosis is a fracture of the femur• The patient has a past history of fractured femur• The patient is suspected of having a fractured femur …. etc

– In a query it may be one of several criteria for retrieving the records of patients with particular types of injury

– In an index to the clinical literature it might indicate a paper that is relevant to this condition

12

Page 13: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Expressions can be pre-coordinated or post-coordinated

• Pre-coordinated expression– Terminology producer provides a single ConceptId for the

meaning• 31978002

– means “fracture of tibia”

• Post-coordinated expression– A user composes a combination of ConceptIds to represent

the meaning• 31978002 : 272741003 = 7771000

– (fracture of tibia : laterality = left)– In human readable form … “fracture of left tibia”

13

Page 14: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

• Refinement

– Replacing value C with a more specific value C1 within an existing (defining) ∃R·C relationship in the definition, giving ∃R·C1

– Example• Fracture of femur

– Defined as: finding-site = bone structure of femur– May be refined to: finding-site = structure of neck of femur

• Yielding the new meaning: Fracture of neck of femur

Refinement and qualification:Two common ways to derive post-coordinated expressions

14

Page 15: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

• Qualification (also called “subtype qualification”)

– Replacing value C with a more specific value C1 within a qualifier ∃R·C relationship (found in the qualifying relationships in the relationships table), giving ∃R·C1

– Example

• Bronchitis– Qualifier exists as: clinical-course = courses (any course value)– May be qualified to: clinical-course = acute (sudden onset

AND/OR short duration)• Yields the meaning: Acute bronchitis

• End results of refinement or qualification are post-coordinated expressions with an identical logical structure

Refinement and qualification:Two common ways to derive post-coordinated expressions

15

Page 16: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Compositional grammar (1)

• Simplest expression is a single conceptid– For example

• 71620000

• Optionally conceptId may be followed by a term enclosed in pipe delimiters– For example

• 71620000|fracture of femur|

• Concepts can be combined with a plus sign that means logical “and” (conjunction)– For example

• 31978002|fracture of tibia| +75591007|fracture of fibula|

16

Page 17: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Compositional grammar (2)

• Refinements can be added after a colonFor example

125605004: 363698007=29627003• Refinements can be nested in parentheses

For example53057004|hand pain|:

363698007|finding site| =(76505004|thumb structure|:272741003|laterality| =7771000|left|)

• Refinements can be grouped in bracesFor example71388002|procedure|:

{260686004|method| =129304002|excision - action|,363704007|procedure site| =66754008|appendix structure|}

Note: the comma also means logical “and” in this expression

17

Page 18: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Severe pain, left thumb

Pain

Finding site

Severity

Thumb structure

Severe

Laterality Left

22253000|pain|:363698007|finding site|=

(76505004|thumb structure|:272741003|laterality|=7771000|left|),

246112005|severity|=24484000|severe|

18

Page 19: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Severe pain, left thumb

Hand Pain

Finding site

Severity

Thumb structure

Severe

Laterality Left

53057004|hand pain|:363698007|finding site|=

(76505004|thumb structure|:272741003|laterality|=7771000|left|),

246112005|severity|=24484000|severe|

19

Page 20: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Subtype relationships

• Every concept is a refined type of one or more other concepts

• For example– “Pain in the leg” is a type of “pain”– “Pain in the leg” is a type of “lower limb

finding”• SNOMED CT represents these defining

relationships with the relationship type called “is a”

20

Page 21: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Subtype relationships

PainLower limb finding

Pain in lower limb

Pain in calf

A pain in the calf is-a pain the lower limb, andPain in the lower limb is-a pain, and is-a lower limb finding

21

Page 22: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Subtype relationships

Infectious diseaseLung disease

Infectious pneumonia

Bacterial pneumonia

Bacterial pneumonia is-a infectious pneumonia, andInfectious pneumonia is-a lung disease, and is-a Infectious disease

22

Page 23: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Why have subtype relationships?

• Because when you selectively retrieve information you usually want to include subtypes

• For example– When searching for “Deep Venous Thrombosis”

you would usually want to retrieve all kinds of DVT including …

• DVT of specific sites (e.g. lower limb)• DVT with particular causes (e.g. air travel related DVT)

… and others

23

Page 24: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Leaf

Root

Deep vein thrombosis of leg related to air travel

Deep venous thrombosis of lower extremity

Deep venous thrombosis

Venous thrombosis

Thrombotic disorder

Vascular disease

Disorder of cardiovascular system

Disorder of body system

Disorder by body site

Finding by site

Clinical finding

Disorder

Disease of vein

Thrombosis of vein of lower limb

Vascular disorder of lower extremity

Disorder of lower extremity

Finding of lower limb

Finding of limb structure

Finding of body region

Disorder of extremity

Peripheral vascular disease

DVT leg assoc w air travel

Subtype hierarchyLooking from leaf to root

24

Page 25: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Other defining relationships

• The difference between two concepts may be represented by other defining relationships– Only relationships that are necessarily

true are defining relationships• For example

– “Pain in calf” has “finding site” “calf structure”

25

Page 26: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Other defining relationships

PainLower limb finding

Pain in lower limb

Pain in calf

A pain in the calf has finding site calfPain in the lower limb has finding site lower limb

lower limbstructure

Calf structure

26

Page 27: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Other defining relationships

Lung diseaseBacterial disease

Bacterial pneumonia

RLL bacterialpneumonia

Bacterial pneumonia has finding site lung structureRLL bacterial pneumonia has finding site RLL structure

Lung structure

RLL structure

27

Page 28: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Why have other defining relationships?

• Other defining relationships– Confirm and enhance the accuracy of the subtype

hierarchy• For example, all “pain” findings with a “finding site” of

“lower limb” (or a subtype of lower limb structure) must be subtypes of “lower limb pain”

– Enable concepts to be refined by increasing the specificity of a defined relationship• For example, a “pain in the foot” could be refined to

specify a more precise “finding site” such as the “third toe of the left foot”, even if SNOMED CT did not include a specific concept for pain in a such a specific location.

– Allow recognition of equivalence between different ways of expressing the same concept

28

Page 29: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Primitive & sufficiently-defined concepts

• A concept is “sufficiently defined” – if its definition is sufficient to

distinguish it from all its supertype concepts

• A concept is “primitive” – if it is not “sufficiently defined”

29

Page 30: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Primitive & sufficiently defined concepts

• Head injury– Is a = Disease– Associated morphology = Traumatic abnormality– Finding site = Head structure– Sufficiently Defined

• Aching pain– Is a = Pain– Primitive

• Headache– Is a = Aching pain– Finding site = Head structure– Sufficiently Defined

30

Page 31: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

The value of sufficiently defining concepts

• Allows auto-classification– Consistent hierarchy and definition

• Allows computation of equivalence and subsumption between– Different ways of expressing the same meaning

• E.g. “open fracture of left femur”

or– “fracture of bone”

site=“femur”: laterality=“left”morphology=“open fracture”

31

Page 32: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Different views of relationships

• Stated view– The view that SNOMED CT modelers edit– Includes only the defining relationships that an

author has explicitly stated to be true– (soon will be distributed in KRSS and/or OWL

syntax)• Inferred view

– The view distributed in the distribution file– Generated by auto-classification– Includes relationships inferred from the stated view– Excludes redundant relationships

• Normalized view– The view best suited to comparing expressions– Reduces all values to their proximal primitive

subtypes

32

Page 33: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Auto-classification

• Many relationships are inferred by auto-classification rather than authored directly

• Auto-classification– Takes definitions “stated” by SNOMED authors and uses

them to “infer” other relationships– Removes redundant (less specific) defining relationships– Creates a logically consistent parsimonious set of

relationships• Review the results of classification

– Although logically consistent … it may not be “correct” due to errors in “stated” definitions

– Human errors that might otherwise be overlooked are often highlighted by auto-classification

– Auto-classification is repeated frequently during authoring and the results are then rechecked

33

Page 34: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

An example of a stated view

pain

pain in lower limb

pain in calf

lower limb structure

calf structure

is a

is a

is a

The diagram is a “directed acyclic graph”, or DAG, of the is-a relationships

34

Page 35: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Auto-classification can addinferred relationships to the stated view

pain

pain in lower limb

pain in calf

lower limb structure

calf structure

Inferredfrom other relationships is a

is a

is a

is a

35

Page 36: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

For the distributed inferred view less specific subtype relationships are removed

pain

pain in lower limb

pain in calf

lower limb structure

calf structure

Redundant after adding inferred

relationship

is a

is a

is a

is a

In computer science terms this structure is called the “transitive reduction”i.e. the distributed is-a relationships are the transitive reduction of the DAG

36

Page 37: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

In another view all the possible subtype relationships are stated directly

• This is “transitive closure” view is useful for optimization

• It can be computed from distributed data

• SNOMED is likely to release this view later this year

pain

pain in lower limb

pain in calf

is a

is a

is a

finding

is ais a

is a

Page 38: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Description Logic Classifiers

• Various DL Classifiers exist• SNOMED uses the Apelon TDE classifier

– It has some limitations but performs well on a large database

• Some other classifiers testing in past failed or became very slow with so many concepts

• Others include– FaCT++ (Fast Classifier of Terminologies)– Pellet– CEL– RacerPro

38

Page 39: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Definition of Normal Forms

• In original RT work, dual independent modeling required exact agreement on stated definition– Resulted in unresolved arguments about modeling style

• State most immediate parent concepts only, and only those relationships that have changed, or

• State proximal primitives only, and all defining relationships

• Defining a normal form allowed different modeling styles for different purposes or preferences

39

Spackman KA. Normal forms for description logic expression of clinical concepts in SNOMED RT. Proceedings/AMIA Annual Symposium. :627-631, 2001.

Page 40: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Illustration of need for normal form:basal cell carcinoma (BCC) of skin of eyelid

• Multiple different ways to post-coordinate:– Disorder, M=BCC, T=skin of eyelid– Malignant disorder, M=BCC, T=skin of eyelid– Skin disorder, M=BCC, T=skin of eyelid– Disorder of skin of eyelid, M=BCC– Malignant neoplastic disorder of skin eyelid,

M=BCC– Basal cell carcinoma (disorder), T=skin of

eyelid– ...

40

Page 41: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

D Disorder

D4Neoplastic disorder of skin

D1Neoplasticdisorder

D3Malignant neoplasm

D2 Disorder of skin

D5 Disorder of skin of eyelid

D7Malignant neoplasm of skin D8 Neoplastic disorder

of skin of eyelidD6 Basal cell

malignancy

D10 Malignant neoplasmof skin of eyelid

D11 Basal cell carcinomaof skin of eyelid

D9Basal cell carcinoma of skin

41

Page 42: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

D Disorder

D4Neoplastic disorder of skin

D1Neoplasticdisorder

D3Malignant neoplasm

D2 Disorder of skin

D5 Disorder of skin of eyelid

D7Malignant neoplasm of skin D8 Neoplastic disorder

of skin of eyelidD6 Basal cell

malignancy

D10 Malignant neoplasmof skin of eyelid

D11 Basal cell carcinomaof skin of eyelid

D9Basal cell carcinoma of skin

NeoplasticMorphology

MalignantMorphology

BCCMorphology

M3

M2

M1T1

T2

Skin structure

Skin structure of eyelid

42

Page 43: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

D Disorder

D4Neoplastic disorder of skin

D1Neoplasticdisorder

D3Malignant neoplasm

D2 Disorder of skin

D5 Disorder of skin of eyelid

D7Malignant neoplasm of skin D8 Neoplastic disorder

of skin of eyelidD6 Basal cell

malignancy

D10 Malignant neoplasmof skin of eyelid

D11 Basal cell carcinomaof skin of eyelid

D9Basal cell carcinoma of skin

NeoplasticMorphology

MalignantMorphology

BCCMorphology

M3

M2

M1T1

T2

Skin structure

Skin structure of eyelid

43

Page 44: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

D Disorder

D4Neoplastic disorder of skin

D1Neoplasticdisorder

D3Malignant neoplasm

D2 Disorder of skin

D5 Disorder of skin of eyelid

D7Malignant neoplasm of skin D8 Neoplastic disorder

of skin of eyelidD6 Basal cell

carcinoma

D10 Malignant neoplasmof skin of eyelid

D11Basal cell carcinomaof skin of eyelid

D9Basal cell carcinoma of skin

NeoplasticMorphology

MalignantMorphology

BCCMorphology

M3

M2

M1T1

T2

Skin structure

Skin structure of eyelid

44

Page 45: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

D Disorder

D4Neoplastic disorder of skin

D1Neoplasticdisorder

D3Malignant neoplasm

D2 Disorder of skin

D5 Disorder of skin of eyelid

D7Malignant neoplasm of skin D8 Neoplastic disorder

of skin of eyelidD6 Basal cell

malignancy

D10 Malignant neoplasmof skin of eyelid

D11 Basal cell carcinomaof skin of eyelid

D9Basal cell carcinoma of skin

NeoplasticMorphology

MalignantMorphology

BCCMorphology

M3

M2

M1T1

T2

Skin structure

Skin structure of eyelid

Note that D1 through D11 are all sufficiently defined (non-primitive)45

Page 46: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

D Disorder

D4Neoplastic disorder of skin

D1Neoplasticdisorder

D3Malignant neoplasm

D2 Disorder of skin

D5 Disorder of skin of eyelid

D7Malignant neoplasm of skin D8 Neoplastic disorder

of skin of eyelidD6 Basal cell

malignancy

D10 Malignant neoplasmof skin of eyelid

D11 Basal cell carcinomaof skin of eyelid

D9Basal cell carcinoma of skin

NeoplasticMorphology

MalignantMorphology

BCCMorphology

M3

M2

M1T1

T2

Skin structure

Skin structure of eyelid

Attributes and values are all inheriteddownwards (redundant ones are removed)

M=M1

M=M2

M=M3

T=T1

T=T2M=M1T=T1

M=M1T=T2

M=M2T=T1

M=M2T=T2

M=M3T=T1

M=M3 T=T246

Page 47: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

D Disorder

D4Neoplastic disorder of skin

D1Neoplasticdisorder

D3Malignant neoplasm

D2 Disorder of skin

D5 Disorder of skin of eyelid

D7Malignant neoplasm of skin D8 Neoplastic disorder

of skin of eyelidD6 Basal cell

malignancy

D10Malignant neoplasmof skin of eyelid

D11 Basal cell carcinomaof skin of eyelid

D9Basal cell carcinoma of skin

NeoplasticMorphology

MalignantMorphology

BCCMorphology

M3

M2

M1T1

T2

Skin structure

Skin structure of eyelid

Some valid forms for D11:

D9 and T=T2

D10 and M=M3

D6 and D547

Page 48: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

D Disorder

D4Neoplastic disorder of skin

D1Neoplasticdisorder

D3Malignant neoplasm

D2 Disorder of skin

D5 Disorder of skin of eyelid

D7Malignant neoplasm of skin D8 Neoplastic disorder

of skin of eyelidD6 Basal cell

malignancy

D10Malignant neoplasmof skin of eyelid

D11 Basal cell carcinomaof skin of eyelid

D9Basal cell carcinoma of skin

NeoplasticMorphology

MalignantMorphology

BCCMorphology

M3

M2

M1T1

T2

Skin structure

Skin structure of eyelid

Some valid forms for D11:

D9 and T=T2

D10 and M=M3

D6 and D548

Page 49: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

D Disorder

D4Neoplastic disorder of skin

D1Neoplasticdisorder

D3Malignant neoplasm

D2 Disorder of skin

D5 Disorder of skin of eyelid

D7Malignant neoplasm of skin D8 Neoplastic disorder

of skin of eyelidD6 Basal cell

malignancy

D10Malignant neoplasmof skin of eyelid

D11 Basal cell carcinomaof skin of eyelid

D9Basal cell carcinoma of skin

NeoplasticMorphology

MalignantMorphology

BCCMorphology

M3

M2

M1T1

T2

Skin structure

Skin structure of eyelid

Some valid forms for D11:

D9 and T=T2

D10 and M=M3

D6 and D549

Page 50: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Summary of BCC example:

• Because of the many different valid forms, it is useful to define a “normal form” to which we can transform all expressions for comparison

• In this case, the normal form combines the proximate primitive (disorder) with the non-redundant existential restrictions

disorder :finding-site = skin of eyelid (body structure),associated-morphology = basal cell carcinoma (morphologic abnormality)

50

Page 51: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

CONCEPT MODEL

51

Page 52: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Procedure site

Direct morphology

Direct substance

Procedure device

Indirect device

Using device

Anatomical structure, Acquired body structure

Substance, Pharmaceutical / biologic product

Procedure concept model (1)

Indirect morphology

Procedure Morphology

Morphologically abnormal structure

Direct deviceDevice

Method Action

Procedure

Procedure site - direct

Procedure site - indirect

Using access device Endoscope and subtypes

52

Page 53: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Procedure site

Direct morphology

Direct substance

Procedure device

Indirect device

Using device

Anatomical structure, Acquired body structure

Substance, Pharmaceutical / biologic product

Procedure concept model (1)

Indirect morphology

Procedure Morphology

Morphologically abnormal structure

Direct deviceDevice

Method Action

Procedure

Procedure site - direct

Procedure site - indirect

Using access device Endoscope and subtypes

DirectObject

53

Page 54: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Has intent

Recipient category

Revision status

Procedural approaches

Priorities

Clinical findings, Procedures

Intents

Person, Family, Community, Donor, Group

Revision-value, Part of multistage procedure, Primary operation

Surgical Approach

Priority

Has focus

Procedure concept model (2)

Route of administration Route of administration value

Physical forcesUsing energy

Substances

Procedure

Using substance

Access Open, closed, percutaneous(surgical access values)

54

Page 55: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Procedures (1)

• Method– The action being performed to accomplish the procedure

• Does not include access (e.g., percutaneous) or approach (e.g., translumbar)– Values are subtypes of “action”

• Procedure site• Values are subtypes of “anatomical concept”

– Procedure site – direct• The site directly affected by the action

– Procedure site – indirect• A site that is acted on but is not the direct target of the action

appendectomy

Gmethod excision - action

procedure site - direct appendix structure55

Page 56: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Procedures (2)

• Morphology• Values are subtypes of “morphologically abnormal structure”

– Direct morphology• The morphology to which the procedure is directed

– Indirect morphology• A morphology that is acted upon, but is not the direct target of

the action being performed

Evacuation of intracerebral hematoma

Gmethod evacuation - action

procedure site - indirect cerebral structuredirect morphology hematoma

56

Page 57: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Procedures (3)

• Device• Values are subtypes of “device”

– Direct device• The device to which the procedure is directed.

– Indirect device• A device that is acted upon, but which is not the direct

target of the action being performed

Replacement of electronic heart device battery

direct device pacemaker battery

indirect device cardiac pacemaker

57

Page 58: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Procedures (4)

• Using device– Specifies the instrument utilized to perform the procedure– Value are subtypes of “device”

core needle biopsy of larynx

Using device core biopsy needle

Using access device

– Specifies the instrument utilized to gain access to the site– Value are subtypes of “endoscope”

Endoscopic biopsyUsing access device endoscope

58

Page 59: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Procedures (5)

• Using energy– Specifies the energy utilized to perform the procedure– Value are subtypes of “physical force”

gamma ray therapy

Using energy gamma radiation

59

Page 60: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Procedures (6)

• Access– Used to distinguish open, closed, endoscopic, and percutaneous

procedures• Only used if access is specified in the name of the concept

– Values are subtypes of “surgical access values”

transurethral laser prostatectomySurgical approach transurethral approach

open reduction of fracture

access open approach

Surgical Approach– Specifies the directional, relational or spatial access to the

site of a surgical procedure. – Values are subtypes of “surgical approach”

60

Page 61: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Measurement procedure concept model

Component

Measurement method

Time Aspect

Scale Type

Cell structure, organism, substance, observable

Laboratory procedure categorized by method

Time frame

Property Property of measurement

Nominal, narrative, ordinal,Quantitative, qualitative, Text, ordinal OR quantitative

Has Specimen Specimen

MeasurementProcedure

61

Page 62: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Measurement procedures

• Has specimen– Specifies the type of specimen on which a measurement or

observation is performed– Values are subtypes of “specimen”

• Component– Specifies the substance or observable being observed or

measured by a procedure– Values are subtypes of “substance”, “observable entity”, or “cell

structure”

creatinine measurement, 24 hour urinehas specimen 24 hour urine sample

component creatinine

62

Page 63: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding concept model (1)

Finding site

Causative agent

Associated morphology

Interprets

Acquired body structure, Anatomical concepts

Organism, Substance, Physical object, Physical force

Morphologically abnormal structure

Laboratory procedure, Observable entity, Patient evaluation procedure

Due to

After Clinical finding, Procedure

Clinical finding, Event

Has interpretation Findings values, Result comments

Associated withClinical finding, Substance, Physical object, Physical force, Events, Organisms, Pharmacological / Biological product, Procedure

Clinical finding

63

Page 64: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding concept model (2)

Has definitional manifestation

Occurrence

Pathological process

Clinical finding

Periods of life

Clinical Course

Episodicity

Courses

First episode, New episode, Ongoing episode

Severity

Finding method

Pathological process

Mild, Moderate, SevereClinical finding

Finding informer

Procedure

Performer of method, Subject of recordProvider of history other than subject,Subject of record or other provider of history

64

Page 65: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Clinical Findings (1)

• Associated morphology– Specifies morphologic change seen at tissue or cellular level as

a characteristic feature or the disease– Values are subtypes of “morphologically abnormal structure”

• Finding site– Specifies the body site affected by a condition– Values are subtypes of “anatomical concept”

open fracture of femur

associated morphology fracture, open

finding site bone structure of femur

65

Page 66: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Clinical Findings (2)

• Causative agent– Specifies the direct causative agent of the disease

• Does not include vectors (e.g. not mosquito for malaria)

– Values are subtypes of “organisms”, “substances”, “physical objects” or “physical forces”

bacterial pneumonia

causative agent bacteria

66

Page 67: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Clinical Findings (3)

• Due to– Used when one finding/disorder is a cause of another finding, disorder

or procedure. Differs from Causative agent in that the cause is not an Organism, Substance, Physical Object, or Physical agent

– Values are subtypes of “clinical findings”

post-viral disorderafter viral disease

• After– Specifies a temporal relationship between a Finding/Disorder and a

Finding, Disorder, or Procedure when there is not necessarily a causal relationship

– Values are subtypes of “clinical findings” or “procedures”

diabetic retinopathy

due to diabetes mellitus

67

Page 68: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Clinical Findings (4)

acute myocardial infarctionClinical course acute

Clinical Course– Specifies the course of a condition. – Used for acute and chronic conditions. – Not used to refer to rapidity of onset or severity of a

condition. – Values are subtypes of “courses”

68

Page 69: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Clinical Findings (5)

• Episodicity– Specifies the particular episode of a finding that may recur– If an episode is not the initial episode and is not an ongoing

episode, it is considered a new episode– Values are “first episode”, “new episode” & “ongoing episode”

• Severity– Specifies the level of severity for a Disease concept. – Values are “mild”, “moderate”, and “severe”

severe vertigo

severity severe

new onset angina

episodicity new episode

69

Page 70: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Attributes Clinical Findings (6)

• Interprets– Specifies the “observable entity” or “function” being evaluated or

interpreted by a finding.– Values are subtypes of “observable entity”, “biological function” or

“measurement procedure”• Has interpretation

– Specifies the judgement being made about an observable or function (e.g., presence, absence, degree, normality, etc.)

– Values are subtypes of “findings values” or “result comments”

decreased cardiac output

has interpretation decreasedinterprets cardiac output

70

Page 71: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Associated procedure Procedure

Procedure contextContext values for actions• Done, not done• Planned, requested

Associated findingClinical finding; or Observable / Observation with result

Finding context

Finding context value• Present, absent, possible• Unknown• Goal, risk, etc

Situation concept model

Situation with explicit context

Subject relationship contextSubject relationship value• Subject of record• Family member, etc

Temporal context Temporal context value• Current• Past, etc

71

Page 72: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Specimen concept model

Specimen Specimen source morphology Morphology

Specimen substance

Specimen source identity Person, Family, Donor, Device, Environment, Community

Substance

Specimen source topography

Specimen procedure

Body structure

Procedure

72

Page 73: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Pharmaceutical/Biologic Product concept model

Has dose form

Pharmaceutical / Biologic product Type of drug preparation (product)

Has active ingredient Substance (substance)

73

Page 74: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Body structure concept model

Left, Right, Right and left, (unilateral)Body structure Laterality

Part-of Body structure

Note: use of “unilateral” implies one side and not the other.This is a type of negation, and therefore unilateral procedures and unilateral findingsactually must be in the situation hierarchy.

74

Page 75: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

RIGHT IDENTITIES

75

Page 76: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Right identity(restricted role value maps)

• R ° S ⊑ R

• x Ry y Sz → x Rz

• R is causative agent• S is has active ingredient

• allergyToAspirin v ∃ causativeAgent.aspirinSubstance• aspirinProduct v ∃ hasActiveIngredient.aspirinSubstance• allergyToAspirinProduct v ∃ causativeAgent.aspirinProduct

• Allows the automated inference that:

– allergyToAspirinProduct v allergyToAspirin

76

Page 77: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Right identity(restricted role value maps)

• R ° S ⊑ R

• x Ry y Sz → x Rz

• R is finding-site• S is part-of

• femurFracture v ∃ site.femur• headOfFemurFracture v ∃ site.headOfFemur• headOfFemur v ∃ part-of.Femur

• Allows the automated inference that:– headOfFemurFracture v FemurFracture

• But this isn’t the purpose for which we use right identity in the current release !

77

Page 78: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Avoiding Right Identities by Using SEP Triplets

Liver Structure

Entire liverLiver Part

XM0Ps Liver structureT-62000 Liver

T-D0535 Liver part 7N330 Liver

Lobe of liver

78

Page 79: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

ROLE GROUPS

79

Page 80: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Role Groups

• Certain defining relationships may be grouped to indicate the way they relate to each other

• For example:– Laparoscopy inspects the peritoneal cavity– Appendicectomy excises the appendix– Laparoscopic appendicectomy

• inspects the peritoneal cavityand• excises the appendixbut does not• Excise the peritoneal cavity

– Grouping action and site avoids misinterpretation

80

Page 81: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Role groups

• Another example

Method: ExcisionProcedure site: Gallbladder

Method: ExplorationProcedure site: Bile duct

Cholecystectomy and exploration of bile duct

Role group 1

Role group 2

81

Page 82: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Role Grouping as a Compromise

• Implementers and modelers fear/loathe nesting of expressions– Nesting violates simple flat frame-based model

• Reality demands faithful representation

• Role grouping attempted (with partial success) to hide the complexity– But it was misunderstood by some in DL community as

being a proprietary hack

82

Spackman KA, Dionne R, Mays E, Weis J. Role grouping as an extension to the description logic of Ontylog motivated by concept modeling in SNOMED. Proceedings/AMIA Annual Symposium. :712-716, 2002.

Page 83: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Need for Role Groups

• When a single concept may have more than one value for a particular attribute– for example, “bone fusion with tendon transfer”

• method = fusion, site = bone, and• method = transfer, site = tendon

• And, one attribute-value pair needs to be associated with another.– How can we specify that the fusion is done to the bone and

not to the tendon? and that the transfer is done to the tendon and not to the bone?

83

Page 84: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Role Groups as a Solution

• Informally: – don’t nest or create sub procedures– simply “group” the attribute-value pairs

• Using curly braces as a syntactic marker:{ site=bone, method=fusion}, {site=tendon, method=transfer}

• Or, in tabular form, use a “group” column:attr value groupsite bone 1method fusion 1site tendon 2method transfer 2

84

Page 85: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Role Grouping Logical Form: A Nested Existential Restriction

• C ⊑ ∃ RRG .(∃R1.C1 ⊓ ∃R2.C2) ⊓ ∃ RRG .(∃R3.C3 )

• Distributed as three 4-tuples in relationships table:C R3 C3 0

C R1 C1 1

C R2 C2 1

– Role group numbers are arbitrary integers, and not designed to be stable across changes in the concept definition

85

Page 86: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

ROLE HIERARCHIES

86

Page 87: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Role (attribute) hierarchies

• Selected SNOMED CT attributes have a hierarchical relationship to one another known as “role hierarchies.” In a role hierarchy, one general attribute is the parent of one or more specific subtypes of that attribute. Concepts defined using the more general attribute can inherit concepts modeled with the more specific subtypes of that attribute.

87

Page 88: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Role hierarchies – procedures

• PROCEDURE DEVICE – DIRECT DEVICE – INDIRECT DEVICE – USING DEVICE – USING ACCESS DEVICE

• PROCEDURE MORPHOLOGY – DIRECT MORPHOLOGY – INDIRECT MORPHOLOGY

• PROCEDURE SITE – PROCEDURE SITE - DIRECT – PROCEDURE SITE - INDIRECT

88

Page 89: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Role hierarchies – clinical findings

• ASSOCIATED WITH role hierarchy: • ASSOCIATED WITH

– AFTER – DUE TO – CAUSATIVE AGENT

89

Page 90: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

SNOMED version

Concept & Role-forming Operators

Role axioms

Language Role grouping

Early work (1996-1999)

(⊓, ∃R:C)( ) EL No

SNOMED RT (2000-2001)

(⊓, ∃R:C)(+) EL+ No

SNOMED CT (Jan02-Jan04)

(⊓, ∃R:C)( ) EL Yes

SNOMED CT (Jul04-present)

(⊓, ∃R:C)(+) R⊑ S ELH+ Yes

Summary of SNOMED’s use of DL

90

Notation mostly follows Donini in Ch.3 Description Logic Handbook(+) means right identities were used

Page 91: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

CONTEXT MODEL

91

Page 92: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Common patterns

• Caveat: these are intended for illustrative purposes only, as examples of ways that system builders might simplify post-coordination for their clinical experts

• They are full logical models for which the meaning can be represented in the interface and in storage with more than one split between the information model and terminology model

92

Page 93: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Common patterns

• Clinical finding present• Clinical finding absent• Clinical finding unknown• History of• No history of• Family history of• No family history of• Observable + value• Procedure done• Procedure not done• (Drug or procedure) contraindicated

• Plus:– all the above with site, or site & laterality

93

Page 94: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding present

SituationAssociated finding <finding>

Finding context Known present

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

94Clinical-finding-present (<finding>)

Page 95: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Abrasion of upper limb

SituationAssociated finding Abrasion of upper limb

Finding context Known present

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

95Clinical-finding-present (abrasion of upper limb)

Page 96: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding absent

SituationAssociated finding <finding>

Finding context Known absent

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

96Clinical-finding-absent (<finding>)

Page 97: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

No chest retractions

SituationAssociated finding Chest wall retraction

Finding context Known absent

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

97Clinical-finding-absent (chest wall retraction)

Page 98: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding unknown

SituationAssociated finding <finding>

Finding context Unknown

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

98Clinical-finding-unknown (<finding>)

Page 99: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Splenomegaly: unknown

SituationAssociated finding splenomegaly

Finding context Unknown

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

99Clinical-finding-unknown (splenomegaly)

Page 100: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

History of <finding>

SituationAssociated finding <finding>

Finding context Known present

Temporal context Current

Group

Subject relationship context Subject of record

In the past

100History-of (<finding>)

Page 101: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

History of MI

SituationAssociated finding Myocardial

infarction

Finding context Known present

Temporal context Current

Group

Subject relationship context Subject of record

In the past

101History-of (myocardial infarction)

Page 102: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

No history of <finding>

SituationAssociated finding <finding>

Finding context Known absent

Temporal context Current

Group

Subject relationship context Subject of record

All times past

102No-history-of (<finding>)

Page 103: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

No history of seizure

SituationAssociated finding seizure (finding)

Finding context Known absent

Temporal context Current

Group

Subject relationship context Subject of record

All times past

103No-history-of (seizure (finding) )

Page 104: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Family history of <finding>

SituationAssociated finding <finding>

Finding context Known present

Temporal context Current

Group

Subject relationship context Family member

In the past

104Family-history-of (<finding>)

Page 105: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Family history of ischemic heart disease

SituationAssociated finding Ischemic heart

disease

Finding context Known present

Temporal context Current

Group

Subject relationship context Family member

In the past

105Family-history-of (ischemic heart disease)

Page 106: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

No family history of <finding>

SituationAssociated finding <finding>

Finding context Known absent

Temporal context Current

Group

Subject relationship context Family member

All times past

106No-family-history-of (<finding>)

Page 107: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

No family history of dementia

SituationAssociated finding dementia

Finding context Known absent

Temporal context Current

Group

Subject relationship context Family member

All times past

107No-family-history-of (dementia)

Page 108: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Switch to procedures

• Slightly different:– Two attributes:

• Associated-procedure• Procedure-context

• Same:– Temporal context– Subject relationship context

108

Page 109: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Procedure done

SituationAssociated procedure <procedure>

Procedure context Done

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

109Procedure-done (<procedure>)

Page 110: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Tetanus booster given

SituationAssociated procedure

Booster tetanus vaccination (procedure)

Procedure context Done

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

110Procedure-done (booster tetanus vaccination)

Page 111: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Procedure not done

SituationAssociated procedure <procedure>

Procedure context Not done

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

111Procedure-not-done (<procedure>)

Page 112: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Neurological examination not done

SituationAssociated procedure

Neurological examination (procedure)

Procedure context Not done

Temporal context Current

Group

Subject relationship context Subject of record

Current or specified time

112Procedure-not-done (neurological examination)

Page 113: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Current or specified time

Drug contraindicated

Situation Associated procedure

<substance>

Procedure context Contraindicated

Temporal context

Group

Subject relationship context Subject of record

Administrationof medication

Direct-substance

113Drug-contraindicated (<substance>)

Page 114: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Current or specified time

Warfarin contraindicated

Situation Associated procedure

warfarin

Procedure context Contraindicated

Temporal context

Group

Subject relationship context Subject of record

Administrationof medication

Direct-substance

114Drug-contraindicated (warfarin)

Page 115: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Add site to previous patterns

• Clinical finding present + site• Clinical finding present + site + laterality• Clinical finding absent + site• Clinical finding absent + site + laterality

115

Page 116: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding present + site

Situation

Associated finding

Finding context

<finding>

Known present

Finding-site <site>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

116Clinical-finding-present-with-site (<finding>,<site>)

Page 117: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Bleeding finger

Situation

Associated finding

Finding context

bleeding

Known present

Finding-site Finger structure

Group

Temporal context Current or specified time

Subject relationship context Subject of record

117Clinical-finding-present-with-site (bleeding, finger structure)

Page 118: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Bleeding index finger

Situation

Associated finding

Finding context

bleeding

Known present

Finding-site Index finger structure

Group

Temporal context Current or specified time

Subject relationship context Subject of record

118Clinical-finding-present-with-site (bleeding,index finger structure)

Page 119: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding absent + site

Situation

Associated finding

Finding context

<finding>

Known absent

Finding-site <site>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

119Clinical-finding-absent-with-site (<finding>,<site>)

Page 120: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

No right femoral bruit

Situation

Associated finding

Finding context

Femoral bruit

Known absent

Finding-site Right femoral artery

Group

Temporal context Current or specified time

Subject relationship context Subject of record

120Clinical-finding-absent-with-site (femoral bruit,right femoral artery)

Page 121: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

No right femoral bruit

Situation

Associated finding

Finding context

bruit

Known absent

Finding-site Right femoral artery

Group

Temporal context Current or specified time

Subject relationship context Subject of record

121Clinical-finding-absent-with-site (bruit,right femoral artery)

Page 122: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding present + site + side

Situation

Associated finding

Finding context

<finding>

Known present

Finding-site <site>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality <side>

122Clinical-finding-present-with-site-and-side (<finding>,<site>,<side>)

Page 123: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Right femoral bruit present

Situation

Associated finding

Finding context

bruit

Known present

Finding-site Femoral artery

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality Right

123Clinical-finding-present-with-site-and-side (bruit, femoral artery, right)

Page 124: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding present + site + side

Situation

Associated finding

Finding context

<finding>

Known present

Finding-site <site>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality <side>

124Clinical-finding-present-with-site-and-side (<finding>,<site>,<side>)

Page 125: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Bleeding of left index finger present

Situation

Associated finding

Finding context

bleeding

Known present

Finding-site index finger structure

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality left

125Clinical-finding-present-with-site-and-side (bleeding, index finger structure, left)

Page 126: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Bleeding skin, left index finger

Situation

Associated finding

Finding context

bleeding

Known present

Finding-site Skin of index finger

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality left

126Clinical-finding-present-with-site-and-side(bleeding, skin of index finger, left)

Page 127: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical finding absent + site + side

Situation

Associated finding

Finding context

<finding>

Known absent

Finding-site <site>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality <side>

127Clinical-finding-absent-with-site-and-side (<finding>,<site>,<side>)

Page 128: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

No right femoral bruit

Situation

Associated finding

Finding context

bruit

Known absent

Finding-site Femoral artery

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality right

128Clinical-finding-absent-with-site-and-side (bruit, femoral artery, right)

Page 129: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Using observables

• Finding present + observable + value

129

Page 130: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Finding present, observable + value

Situation Associated finding

Finding context

Clinical finding

Known present

Has-interpretation <value>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Group

Interprets <observable>

130Finding-present-observable-value (<observable>,<value>)

Page 131: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Knee jerk reflex 2+ (out of 4)

Situation Associated finding

Finding context

Clinical finding

Known present

Has-interpretation ++

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Group

Interprets Knee jerk reflex271714006

260348001

131 Finding-present-observable-value (knee jerk reflex,++)

Page 132: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Situation Associated finding

Finding context

Clinical finding

Known present

Has-interpretation <value>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Group

Interprets <observable>

obs-site <site>

Finding present, observable + site + value (assuming we approve a site attribute for observables)

132Finding-present-obs-site-value (<observable>,<site>,<value>)

Page 133: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Situation Associated finding

Finding context

Clinical finding

Known present

Has-interpretation ++

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Group

Interprets Knee jerk reflex

obs-site Left knee

Left knee jerk reflex ++(assuming we approve a site attribute for observables)

133Finding-present-obs-site-value (knee jerk reflex, left knee,++)

Page 134: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Procedure patterns

• Procedure done + method + site• Procedure done + method + site + laterality

134

Page 135: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Procedure done, method+site

Situation Associated procedure

Procedure context

procedure

Done

Procedure site - direct <site>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Group

Method <method>

135Procedure-done-plus-method-site (<method>,<site>)

Page 136: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

X-ray of wrist done

Situation Associated procedure

Procedure context

procedure

Done

Procedure site - direct wrist

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Group

Method Radiographic imaging

136Procedure-done-plus-method-site (x-ray, wrist)

Page 137: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Procedure done, method+site+side

Situation Associated procedure

Procedure context

procedure

Done

Procedure site - direct <site>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality <side>

Group

Method <method>

137Procedure-done-method-site-side (<method>,<site>,<side>)

Page 138: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

X-ray of left wrist done

Situation Associated procedure

Procedure context

procedure

Done

Procedure site - direct wrist

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality left

Group

Method radiographic imaging

138Procedure-done-method-site-side(radiographic imaging, wrist, left)

Page 139: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

More specific patterns also possible

• For example:– Suture of skin done + site + laterality

139

Page 140: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Suture of skin done, site + side Situation Associated

procedure

Procedure context

Closure of skin by suture

Done

Procedure-site-indirect <site>

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality <side>

Group

Procedure-morphology-direct Laceration

Method Closure by device

Using-device Suture

140

Page 141: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Suture of laceration of skin of left index finger: done

Situation Associated procedure

Procedure context

Closure of skin by suture

Done

Procedure-site-indirect

Skin of index finger

Group

Temporal context Current or specified time

Subject relationship context Subject of record

Laterality left

Group

Procedure-morphology-direct Laceration

Method Closure by device

Using-device Suture

141

Page 142: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

REDESIGN PROJECTS

142

Page 143: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Anatomy Redesign

• Reintroduce part-of roles• Sufficiently define the S and P of the SEP triad • Align the E with the Foundational Model of Anatomy

(FMA)

143

Page 144: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Reflexive roles

• Plan to introduce reflexive “part-of” as a way of handling “SEP” model evolution

proper-part-of v part-of ² v part-ofS ≡ ∃ part-of . EP ≡ ∃ proper-part-of . E

144

Suntisrivaraporn B, Baader F, Schulz S, and Spackman K. Replacing SEP-Triplets in SNOMED CT using Tractable Description Logic Operators. In Jim Hunter Riccardo Bellazzi, Ameen Abu-Hanna, editor, Proceedings of the 11th Conference on Artificial Intelligence in Medicine (AIME'07), Lecture Notes in Computer Science. Springer-Verlag, 287-291, 2007.

Page 145: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Substance Redesign

• Remove inappropriate is-a relationships• Add new attributes and values

• Many difficult decisions remain– How to represent “may be used as …”

• Timolol may be used as an eye medicine for glaucoma• Timolol may be used as a cardiac beta blocker

– Are H2CO3 and HCO3- the same or different? Do we

represent the various anions of drugs also?

145

Page 146: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Organism redesign

• Major steps:1. organize SNOMED's taxonomy into a

systematic and consistent Linneanhierarchy

2. remove all non-taxonomic information about living organisms from the taxonomic hierarchy

3. represent such information, when understandable, reproducible and useful, elsewhere in the terminology

146

Page 147: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

• Linnean taxonomic terms (“Canis familiaris”)

• Common names for organisms (“Dog”)• Non-taxonomic information

– Use and Circumstances • Laboratory fur-bearing animal

– Pathogenicity• Parasite, pyogenic bacterium

– Life cycle stage of organisms • Worm eggs

Current taxonomy includes:

147

Page 148: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Common names in the FSN:

• Some organisms have many common names– Butorides virescens = green heron, green-backed heron,

little green heron, crab-catcher, fly-up-the-creek, green bittern, poke, shitepoke, skeow, skow, and swamp squaggin

• May be impossible to verify what organism is meant– Ex: Comte de Paris star frontlet (organism) ???

• A single common name may refer to more than one species:– Ex: Yellowhammer (organism) MAY BE A Emberiza

citrinella, MAY BE A Colaptes auratus

148

Page 149: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Non-taxonomic terms in a taxonomic hierarchy:

• a subtype is always and necessarily a "kind of" its parent (this is what subsumption means)

• interpolation of non-taxonomic terms in a taxonomic hierarchy violates this convention– these terms are often context-dependent rather than

defining. • An elephant may be a domestic animal in India• a dog may be a food animal in Korea• Is a canary a “Wild bird--chordate” or a “Domestic

fowl”? – Answer: Neither. It is Serinus canaria

149

Page 150: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Further enhancements of the model

• Attributes & values to represent contextual information about living organisms– Contexts of domesticity (domestic, feral, wild)– Contexts of use (food, laboratory, companion, service,

breeding, etc)– Contexts of life stage (oocyst, larva, spore, trophozooite,

etc)– Contexts of medical significance (parasite, renotrophic

organism, pathogen)???

150

Page 151: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Qualifiers for organisms?

• An organism might be qualified by non-taxonomic attributes/values, just as a disease might be qualified by severity, stage, episode, degree of control, etc.

• But: – Type 2 diabetes that is out of control is not really a different

type of disease; it is a different type of situation in which type 2 diabetes (the disease) is present.

– Dairy cattle and beef cattle are not really different types of organisms; they are different types of contexts in which cattle (organisms) are used for different purposes.

151

Page 152: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

What about “infectious agents”?

• The taxonomy of parasites, bacteria and other potentially pathogenic microorganisms is also a mixture of scientific names, common names, and contextual information

• Attempting to convey “contexts of pathogenicity” creates errors in logic:– Ex: Helminth ISA Parasite?

• Wrong. Most helminths are not parasitic– Ex: Fungus ISA Infectious agent?

• Wrong. Most fungi are not infectious

152

Page 153: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Observable Redesign

• Separate processes, functions, and qualities

• Add attributes that define observables in terms of:– Properties they observe– Timing– Scales or units– Techniques of observation

• Add attributes that define qualities/properties in terms of:– the independent continuant in which they inhere

153

Page 154: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Substances,  functions, processes, activities, organisms, cell structures

Observable entity

COMPONENT

PROPERTY

SYSTEM / OBSERVABLE SITE

Properties

Specimens, Sites

TIME ASPECT

SCALE / UNITS

Time aspects

Scale types, units

TECHNIQUE techniques

DRAFT model of observables

154

Page 155: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Observable entity

PROPERTY properties

independent continuant

TIME ASPECT

SCALE / UNITS

Time aspects

Scale types, units

TECHNIQUE techniques

INHERES INTOWARDS Functions, substances

DRAFT alternative model of observables

155

Page 156: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Events, conditions, episodes

• Need to define what is an event, what is a condition, what is an episode

• Need criteria for deciding whether we need one code or two– seizure, epilepsy: clearly different, so we need two codes– tachycardia, tachyarrhythmia: same or different?– low hemoglobin, anemia: same or different?– rash of forearm: do we need both a disorder and a finding?

156

Page 157: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Should we add more expressive DL features?

• General concept inclusion axioms• Transitive roles• Reflexive roles• Disjointness axioms• Value restrictions• Negation• Disjunction• Cyclic definitions• Number restrictions

157

Page 158: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

General concept inclusion axioms

• Extremely useful feature• Compatible with a polynomial-time structural

subsumption algorithm

• Allows us to say what is true in addition to what is sufficient– Gastric ulcer is located in the stomach, and in addition it

necessarily involves the gastric mucosa

158

Page 159: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Transitive roles

• x Ry y Rz → x Rz

• Useful for causal/associational chains• Interaction with role hierarchy is interesting &

useful

• Example: Associated-with-after– Varicella (chicken pox)

– An infection with causative-agent = varicella virus– Herpes zoster

– Also has causative-agent = varicella virus, and occurs after varicella

– Post-herpetic neuralgia– Occurs after herpes zoster (therefore occurs after varicella),

but is not an infection with causative-agent varicella virus

159

Page 160: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Reflexive roles

• Plan to introduce reflexive “part-of” as a way of handling “SEP” model evolution

proper-part-of v part-of ² v part-ofS ≡ ∃ part-of . EP ≡ ∃ proper-part-of . E

160

Suntisrivaraporn B, Baader F, Schulz S, and Spackman K. Replacing SEP-Triplets in SNOMED CT using Tractable Description Logic Operators. In Jim Hunter Riccardo Bellazzi, Ameen Abu-Hanna, editor, Proceedings of the 11th Conference on Artificial Intelligence in Medicine (AIME'07), Lecture Notes in Computer Science. Springer-Verlag, 287-291, 2007.

Page 161: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Value restriction ∀R.C

• Not an intuitive construct– person u ∀hasCar.Jaguar– Includes people who have no car, but if they had one it

would have to be a Jaguar . . . . Do we encounter this kind of concept in common-sense thinking?

• Creates pernicious interactions with disjunction and negation that tend to make structural subsumption algorithms incomplete

• But it was included in ALC and FL−, so languages including it were studied extensively.

161

Page 162: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Negation ¬C

• Head injury without loss of consciousness

headInjury u ¬ lossOfConsciousness

situation u∃ includes.headinjury u¬ ∃ includes.lossOfConsciousness

162

Page 163: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Disjunction C t D

• Some high-level aggregators are naturally disjunctive

• We can address this need partially by using navigation hierarchies

163

Page 164: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Cyclic definitions, number restrictions

• ? No significant need for these at present

164

Page 165: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

INFORMATION MODEL INTERACTIONS

165

Page 166: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Clinical Decision Support Model + Inference Rules

Terminology Model+ Compositional Expressions Information Model

+ Patient Data Structures

Diagram based on Figure 1 in Rector AL et al. “Interface of Inference Models with Concept and Medical Record Models” AIME 2001: 314-323

Interaction between Terminology and Information Models

166

Page 167: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Terminology vs Information modelWhat’s the issue?

• Information model:– Determines and organizes the kinds of entities which

carry values in a record– Loosely referred to as slots, facets, fields, questions

• Terminology model:– Determines and organizes the kinds of entities which

are the values– Variously referred to as the terminology or ontology

or value sets

167

Page 168: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Extremes (reductio ad absurdum)

• Put all meaning in the terminology– A code (or terminology expression) for every meaning

that needs to be expressed– Only one “field” in the record

• What about dates, numeric values, names, places, and relationships between them?

• Put all meaning in the information model– Two values: “yes” and “no”– A “question” for every meaning that needs to be

expressed, and a field for every question• What about combinatorial explosion of subtypes of

things in the real world?

168

Page 169: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Representing the semantics of clinical data

• For any given application– There needs to be a boundary between information

model and terminology• without gaps or overlaps

• There are several different choices for where to draw the boundary between

• No single choice of boundary is globally the best• How can we achieve standardization for

interoperability?

169

Page 170: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

Terminology – information model interaction: broad tasks required

• Identify gaps and overlaps• Design a strategy to

– Fill the gaps– Manage the overlaps

• Demonstrate implementability

170

Page 171: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

TermInfo – Specific advice

Act.code & Observation.value (1)• HL7 theory

– code nature of observation– value value of the observation

• Practical implementation– Simple for numeric observations

• Hemoglobin level (code) = 14g/dL (value)– Reasonable for observations with coded results

• Visual acuity (code) = Can count fingers (value)– Tricky for observations where the distinction between

the nature and value is arbitrary• “Blood group AB” could be …

1. ABO Blood grouping (code) Blood group AB (value)

2. Blood group A antigen (code) Present (value)and Blood group B antigen (code) Present

(value)3. Blood group AB (code)

171

Page 172: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

TermInfo – Specific adviceAct.code & Observation.value (2)

• The code-value split is even more arbitrary for general clinical observations

• For example– Finding of abdominal tenderness …

1. Examination (code) abdomen tender (value)2. Abdominal examination (code) abdomen tender

(value)3. Abdominal palpation (code) abdomen tender

(value)4. Abdominal tenderness (code) present (value)5. Abdomen tender (code)

172

Page 173: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

TermInfo – Specific advice

Code and Value – guidance• HL7 code and value distinction should be used for

– Numeric and non-numeric results of measurement procedures

• A single coded attribute should express the full semantics – If there is no non-arbitrary reproducible

distinction– Recommended Implementation

• Act.Code = ASSERTION• Observation.Value = Coded SNOMED

expression• Rationale

– SNOMED CT clinical findings are • not just the value of a particular type of

observation• equivalent to an observable or observation

type with a value

173

Page 174: Knowledge Representation in Practice: SNOMED CT - KR-MED.org

“Code-value” discussion

• Not unique to HL7• Suggests terminology-information model

standardization efforts may benefit from each other

174