SNOMED Clinical Terms: Concepts and Descriptions

Kent A. Spackman, MD, PhD Oregon Health & Science University, Portland OR Chair, SNOMED International Editorial Board

HL7 UK London – November 2004

2

Which concepts and which descriptions?

• • • • • What is SNOMED CT?

What is it for?

How can it be used?

 Relationship to information models, patient data What is involved in developing & maintaining it?

  Semantics Design principles & logic foundation Opportunities for having input

3 `I don't know what you mean by "glory,"' Alice said. Humpty Dumpty smiled contemptuously. `Of course you don't -- till I tell you. I meant "there's a nice knock-down argument for you!"' `But "glory" doesn't mean "a nice knock-down argument,"' Alice objected.

`When I use a word,' Humpty Dumpty said in rather a scornful tone, `it means just what I choose it to mean -- neither more nor less.'

`The question is,' said Alice, `whether you can make words mean so many different things.' `The question is,' said Humpty Dumpty, `which is to be master - - that's all.' From

Through the Looking Glass

, Lewis Carrol

4

What is it?

• “The Systematized Nomenclature of Medicine”

5

What is it?

• • • • • • • • • • A reference terminology A clinical terminology  with reference and interface properties A CD containing a set of tables A set of codes with names A set of definitions “per genus et differentiam” A clinical terminology standard A knowledge base?

A dictionary?

An ontology?

An application ontology?

6

Formal Ontology?

• • SNOMED is not a formal ontology (but some parts of it are migrating in that direction) It is a reference terminology that is progressively more well supported by formal ontological principles   Includes terms and non-ontological assertions / ideas I dislike the term “application ontology” – fish or fowl?

• • Many of SNOMED’s design decisions are supported by formal ontological principles.

 But… Many of SNOMED’s hierarchies are still “unprincipled” and incomplete.

 Requires continued evolution and maturation

7 Formation of the SNOMED International Division of the C.A.P.

SNOMED – CTV3 Timelines SNOMED Read Codes

SNOMED 2 SNOMED 3 CAP business plan

NHS Agreement

SNOMED RT 1979 1980 1981 1982 1983 Read Codes (v1) 1984 1985 1986 1987 1988 Professional Endorsement 1989 1990 Purchased by NHS 1991 1992 Clinical Terms Projects 1993 “ 1994 “ 1995 CTV3 (Clinical Terms version 3) 1996 UK Gov’t Inquiries into Read Codes 1997 “ 1998 “

1999 CAP Agreement

2000 2001 2002

SNOMED Clinical Terms

8

SNOMED CT Releases 1 st 2 nd 3 rd 4 th 5 th 6 th Jan 31, 2002 July 31, 2002 Jan 31, 2003 July 31, 2003 Jan 31, 2004 July 31, 2004 .

.

.

9

Content, Content, Content Number of Concept Codes

350,000 300,000 250,000 200,000 150,000 100,000 50,000 0

10,990 SNOP 1965 30,547 SNM2 1979 96,042 SNM3 1993 Edition of SNOMED 128,030 SRT 2000 325,857 SCT 2002

10

Emergence as a Standard: Recent Events

• Government Actions – US and UK    US National License ANSI – Terminology Distribution Structure Standard US NCVHS – HIPAA recommendation   US Government CHI Initiative recommendation UMLS release  UK NPfIT adoption

11

What does it do?

• SNOMED CT is a terminological resource that can be implemented in software applications to represent clinically relevant information  In a “semantically structured” form that can be used by automated applications

12

What is it for?

• It is for building applications capable of:  Recording statements about the health and health care of individuals • In a way that permits retrieval according to the meaning of the statements, rather than just the words used  Retrieving individual cases and groups of cases • To enable more automated and sophisticated decision support, epidemiology, and research

13

Successful use of SNOMED CT depends on:

• Implementation in clinical records systems • Which in turn requires (at least) a patient data model (information model)

14

The simplest information model

• Put all clinical data here ___________________

The simplest terminology model

• Two values: • Yes • No Intermediate between these extremes there are many possible solutions!

15

What about clinical decision support?

16

What about clinical decision support?

IF Two blood cultures, drawn through an antibiotic removal device, more than 30 minutes apart, grow no organism, THEN discontinue antibiotic.

17 procedures

IF Two

blood cultures , drawn

through

an antibiotic removal device ,

more than 30 minutes apart

,

grow

no organism ,

THEN

discontinue antibiotic .

finding device

Clinical Decision Support Model + Inference Rules 18 Terminology Model + Coded Data 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

Clinical Decision Support Model + Inference Rules IF Two blood cultures , drawn through Antibiotic removal device , more than 30 minutes apart , grows no organism , THEN discontinue antibiotic .

19 SNOMED CT Terminology Model + Coded Data 30088009 blood culture 55512120 antibiotic removal device 264868006 No growth 281789004 antibiotic therapy 223438000 advice to discontinue a procedure HL7 RIM Information Model + Patient Data Structures What test was performed?

How many were done?

At what time?

What device was used?

What was the result of the test?

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

20

What is involved in creating and maintaining SNOMED CT?

• Representation of meaning   Judgments of “same or different” Representing clearly “what clinicians mean when they say …”

21

It is notoriously difficult to tell what people

• •

mean just by what they say

From “The Economist”, Charlemagne column,

Sept 4, 2004

“Decoding a Euro-diplomat takes more than a dictionary”  “Up to a point” means “I agree in part”? Wrong, it means: • “No, not in the slightest”  “I hear what you say” means “He accepts my point of view”? Wrong, it means: • “I disagree and do not want to discuss it any further”     “With the greatest respect” means • “I think you are wrong, or a fool” “By the way” or “incidentally” means “This is not very important”? Wrong, it means • “The primary purpose of our discussion is …” “I’ll bear it in mind”  “I’ll do nothing about it” “Correct me if I’m wrong”  “I’m right, don’t contradict me”

22

Discerning and representing meaning of

•

health terminology is difficult

What is juvenile rheumatoid arthritis?

 Seropositive chronic idiopathic arthritis in child < 16 yrs ?

  Any chronic arthritis in child < 16 yrs?

Is Adult onset Still’s disease included?

• Three different published terminology standards, all incompatible    JRA (juvenile rheumatoid arthritis) – US JCA (juvenile chronic arthritis) – UK JIA (juvenile idiopathic arthritis) – International

23

Words alone are insufficient

• • • • • • • • • • • There are national, regional and local variations in meaning of words and phrases (even within the same language) Multiple meanings with the same “preferred name” Combining words gives something entirely different from the sum of the parts Ambiguous shorthand and abbreviations are common The same phrase means different things to different specialists The same word or phrase means different things depending on what you are doing at the time Significant differences in meaning are often obscured through use of the same word Formal definitions are often at variance with common clinical usage A general name takes on a more specific meaning A manifestation is often used to name the disorder in which it occurs Successful communication relies on making ontological distinctions that are ignored by common phrasing

24

What is “pudding”?

• • • At dinner in Phoenix, Roger (from the UK) asked “Is anyone having pudding?” To which I replied, “Do you mean dessert?” And he said, “No, I mean pudding.” ?

Within the same language there are significant national, regional and local variations

25

What is “scalp”?

• • • scalp: the skin covering the cranium (Stedman’s) scalp: the soft tissue envelope of the cranial vault, consists of 5 layers: the skin, connective tissue, epicranial aponeurosis + occipitofrontalis muscle, loose areolar tissue, and pericranium. (Gardner, Gray & O’Rahilly, anatomy text) Epicranium (Stedman’s): the muscle, aponeurosis and skin covering the cranium

It is quite clear SNOMED must have two different codes (two different Meanings) that bear the name “scalp”

We say like Humpty Dumpty “When

I

use the word scalp, …”

26

What is a “pyogenic granuloma?”

• • Pyogenic = pus forming Granuloma = a collection of inflammatory cells of a particular type • • Pyogenic granuloma = a benign tumor of small blood vessels of the skin It is neither

pyogenic

nor a

granuloma

.

Combinations are frequently very different from the sum of their parts

What is “general paresis”?

• • General = affecting all skeletal muscles Paresis = weakness • GPI = a form of tertiary neurosyphilis characterized by generalized weakness 27 Shorthand and abbreviations are common

28

What is “acute inflammation”?

• To the GP, it is inflammation with an acute onset, characterised by redness, heat, swelling and pain.

• To the pathologist, it is inflammation in which polymorphonuclear leukocytes predominate, as opposed to chronic inflammation, in which “mononuclear cells” (lymphocytes, plasma cells, monocytes, histiocytes) predominate.

The same phrase can mean different things to different specialists

29

What is the “fundus”?

• When caring for a pregnant patient – • When examining the eyes – • When doing a gastroscopy – • When doing a cholecystectomy – What you are doing at the time changes the meaning of words

30

Is there an error in this hierarchy?

Radiographic procedures Angiography procedures Magnetic resonance angiography procedures

Is there an error in this hierarchy?

Radiographic procedures Angiography procedures Magnetic resonance angiography procedures 31 It is common for a general name to acquire a more specific meaning

32

Is there an error in this hierarchy?

psoriasis psoriasis with arthropathy juvenile psoriatic arthritis juvenile psoriatic arthritis without psoriasis

Is there an error in this hierarchy?

psoriasis psoriasis with arthropathy juvenile psoriatic arthritis juvenile psoriatic arthritis without psoriasis 33 It is common for the disorder to be named by its manifestation

34

What is a “laceration”?

• • Torn or jagged wound vs Accidental cut wound • • Perineal

laceration

during O-P delivery vs

Laceration

of thumb while using kitchen knife Subtle distinctions are often implicit

What is the “leg”?

• • 1) same as “lower limb” 2) just the part from the knee to the ankle   Stedman’s “the segment of the inferior limb between the knee and the ankle” Dorland’s “that section of the lower limb between the knee and ankle” 35 Some formal definitions are in conflict with ordinary usage

What does “aspirin” mean?

• • Some

aspirin

– the chemical ASA An

aspirin

– a tablet containing ASA 36 Formal ontologists insist on a clean distinction between the individual and the matter or stuff of which it is made.

37

How does SNOMED address these issues?

• Careful representation of meaning   Evolutionary design Formal description logic foundation  Consensus process • URU criteria: understandable, reproducible, useful

38

We are not the language police

39

Evolutionary Design

• • • Evolution without pre-ordained design Accumulation of desirable features Heterogeneity of perspectives • Dealing with Scale  Participatory consensus-based approach • Involve the experts  Semantics-based concurrency control • Description logic underpinnings  Configuration management tools • Keith Campbell’s “Galapagos” tool set

40

Description Logic Foundation

• • • SNOMED is based on the description logic known as

ELH

 Conjunction   Existential restrictions Role hierarchies Plus “role groups” (see 2002 AMIA paper) Plus role composition  So far, only one: direct-substance o has-active ingredient

41

How large is large?

• With 800,000+ terms in SNOMED CT  if you spent 5 seconds looking at each one it would take you   4 million seconds = 66,666 minutes = 1,111 hours 138 work days if that’s all you did every day  138/5 = almost 28 work weeks • • At SNOMED we don’t just

pretend

to know about the problems of scale.

That’s not saying we think we’ve solved them.

42

Number of attributes (relationship types) in the SNOMED concept model

60 55 50 45 40 35 Jan 02 Jul 02 Jan 03 Jul 03 Jan 04 Jul 04

43

Percentage of SNOMED CT concept codes that are “fully defined” 30% 25% 20% 15% 10% 5% 0% Tot Dis Find Proc Other

Eventually should reach ~70% or more of disorders, findings & procedures

44

How long will it take?

• That depends on what you want:  

It is ready for use now

.

If you wait for perfection you wait forever.

 But tell us what needs the most urgent attention.

45 • • • • • •

SNOMED phases

1975-1994 Roger Cote phase 1995-1997 Kaiser CMT phase 1997-1999 CAP phase – building SNOMED RT 1999-2002 SNOMED – Read merge phase 2002-2004 US/UK endorsement phase 2004- adoption, use & maintenance phase • The hardest part is still ahead

Number of Concept Codes

350,000 300,000 250,000 200,000 150,000 100,000 50,000 0

10,990 30,547 SNOP 1965 SNM2 1979 96,042 128,030 SNM3 1993 SRT 2000 325,857 SCT 2002 Edition of SNOMED

46

There is opportunity to be involved

• • • Open working group meetings + on-line discussion forums Active working groups:  Concept model working group   Mapping working group Content-area focused working groups • Primary care • • • Nursing Genomics Anesthesiology, pathology, dermatology, ophthalmology, … Upcoming in-person meeting dates:  Feb 2, 2005, S. California  June 14-15, 2005, Chicago  Oct 5, 2005, London

47

Concept Model Working Group issues:

• • • • Context Negation Composition (“post-coordination”) Interface between concept model & information model   Specifically interface between SNOMED & HL7 v3 Proposed work item (or possible SIG) with HL7 vocab

48