Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
i n t e r n a t i o n a l j o u r n a l o f m e d i c a l i n f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453
journa l homepage: www. int l .e lsev ierhea l th .com/ journa ls / i jmi
Implications of SNOMED CT versioning
Dennis Leea,∗, Ronald Cornetb, Francis Laua
a School of Health Information Science, University of Victoria, Victoria, BC, Canadab Department of Medical Informatics, University of Amsterdam, Amsterdam, The Netherlands
a r t i c l e i n f o
Article history:
Received 26 February 2010
Received in revised form
18 August 2010
Accepted 16 February 2011
Keywords:
SNOMED CT
Reference terminology
Clinical vocabulary
Versioning
a b s t r a c t
Purpose: To determine the changes each Systematized Nomenclature of Medicine Clinical
Terms (SNOMED CT) release undergoes and the implications of those changes.
Methods: (1) We reviewed the SNOMED CT Component History documentation and analyzed
the Component History table in detail. (2) We outlined a list of semantic changes that are
made to SNOMED CT concepts that are not recorded as part of the Component History
mechanism. (3) We reviewed the SNOMED CT Concept Model mechanism and how it has
changed since the July 2006 release. We applied the above methods to determine the impact
they had on the Clinical Observations Recording and Encoding (CORE) Problem List published
by the United States National Library of Medicine National Institutes of Health.
Results: Of the 5182 concepts in the problem list subset, 2135 (41.2%) underwent some form
of change in the form of (1) fully specified name and preferred term; (2) concept status; (3)
primitive/fully defined status, defining attributes and normal forms; and/or (4) position in
hierarchy (top-level hierarchy, supertypes and subtypes).
Conclusions: In this paper, we have identified four types of changes that occur over time as
new SNOMED CT releases are introduced. Keeping track of these changes is important as
they are not well published and have an impact in patient case queries and the accuracy of
patient records.
status, defining attributes, normal forms, and position withinthe “is a” hierarchy. Some of these changes may lead to unex-
1. Introduction
The Systematized Nomenclature of Medicine Clinical Terms(SNOMED CT) is a reference terminology that covers suchconcepts as diseases, clinical findings and procedures. Orig-inally developed by the College of American Pathologists inthe United States, the ownership of SNOMED CT was trans-ferred to a new public entity called the International HealthTerminology Standards Development Organization (IHTSDO)in 2006. Presently, IHTSDO has 15 charter member countries
with the common goal to develop, maintain and promote thisterminology standard.
∗ Corresponding author at: School of Health Information Science, UniveTel.: +1 250 885 9372.
The July 2009 version of SNOMED CT contains over 388,000concepts, 1.14 million descriptions and 1.38 million relation-ships. There is a new release every six months through theNational Release Centers of the respective charter membercountries. With each release, there are changes that can affectthe use of SNOMED CT within an organization’s electronicpatient record (EPR) systems. These include the fully specifiedname/preferred term, concept status, primitive/fully defined
rsity of Victoria, PO Box 3050, STN CSC Victoria, BC, Canada.
pected consequences in subsequent encoding, equivalencyand subsumption testing, and querying of a SNOMED CT
Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
a l i n
ezCc
nto[otiintfodpco
2
OpHcrrwhptifi
trRUoou
2
Aptc[u(coara
i n t e r n a t i o n a l j o u r n a l o f m e d i c
ncoded EPR system. Therefore, it is important for an organi-ation to review the changes made in each release of SNOMEDT and understand their implications, particularly the con-epts that are part of the organization’s existing subsets.
While there have been previous studies that emphasize theeed for auditing SNOMED CT to improve the quality of theerminology using methods such as area and partial-area tax-nomy [1], formal concept analysis [2], ontological principles
3] and description logic [4], there has been limited discussionn the implications versioning has on healthcare organiza-ions that have implemented SNOMED CT in their clinicalnformation systems. This study builds on some of the knownssues identified with using SNOMED CT as a reference termi-ology and issues with versioning [5–7]. This paper examineshe different mechanisms of tracking changes between dif-erent SNOMED CT versions and their implications on anrganization’s EPR system. We first describe the method andataset used to study the versioning issues involved. We thenresent our findings, and discuss the implications of thesehanges, especially for organizations that are already usingr planning to use SNOMED CT as their terminology standard.
. Method
ur method had three parts. Firstly, we reviewed the Com-onent History documentation and analyzed the Componentistory table in detail. Secondly, we outlined a list of semantichanges that are made to SNOMED CT concepts that are notecorded as part of the Component History mechanism. Weeferred to this as the Semantic History mechanism. Thirdly,e reviewed the SNOMED CT Concept Model mechanism andow it has changed since the July 2006 release. Although aroposal has been made to include a versioning number tohe compositional grammar of SNOMED CT expressions [8], its not included in our method as the details have not beennalized.
We applied these methods to a SNOMED CT subset andracked its changes between the January and July 2009eleases. The subset used was the Clinical Observationsecording and Encoding (CORE) Problem List published by thenited States National Library of Medicine National Institutesf Health. The examples described in this paper are basedn changes made between the January and July 2009 releasesnless otherwise stated.
.1. Component History mechanism
s part of the semi-annual release of SNOMED CT, a Com-onent History table is included. The Component Historyable records the addition, modification and retirement ofoncepts, descriptions, relationships, cross maps and subsets9]. This table contains five columns: (a) ComponentId—Thenique SNOMED CT Identifier for the changed Component;
b) ReleaseVersion—The version of SNOMED CT in which thishange was made; (c) ChangeType—An indication of the nature
f the change. ChangeType can contain one of three values:dded, status change, or minor change. All other changesequire a Component to be inactivated and replaced with
new component with a different SNOMED CT ID; (d)
f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453 443
Status—The status of this Component after the change. Thereare six specialized relationships called historical relation-ships that are used to link inactive concepts with activeconcepts. These historical relationships (“replaced by,” “sameas,” “may be a,” “was a,” “moved to,” and “moved from”)can be used to locate retired, outdated, erroneous, dupli-cate, ambiguous concepts and concepts that have been movedto a different namespace [10]; A namespace refers to anextension to SNOMED CT that an organization creates andis identified using a seven-digit namespace-identifier that isassigned by the IHTSDO. (e) Reason—An optional text descrip-tion of the reason for the change [11]. The history tablegives eight reasons for changes to a SNOMED CT compo-nent: created and retired intra-release, concept status change(only used between January 2003 and July 2004), fully speci-fied name change, description status change, description typechange, initial capital status change, language code change,and status change (refers to concept or description statuschange and is used in every release except when conceptstatus change was used). A reason may also be a combina-tion of these. For example, a change in a description mayinclude a description type change, initial capital status changeand language code change. An example is with the descrip-tion “2694957018|Calcitriol|”, which in January 2008 had thedescription type of “synonym,” initial capital status of “true”and language code of “en-GB” that was changed to “preferred,”“false” and “en” respectively in July 2008. The reasons forchange and frequency between the July 2007 and July 2009releases are shown in Table 1.
The last five releases of SNOMED CT have seen changesranging from 22,584 in the latest July 2009 version to 210,500in the July 2008 version. The number of concepts, descriptionsand relationships, and the total additions, status changes andminor changes in these releases are shown in Table 2. Thechanges to the core tables refer to the addition or deletionof records in the table. Concepts and descriptions cannot beremoved but the status can be changed to inactivate the con-cept or description. On the other hand, if relationships areno longer valid, they can be removed, which explains whythere are sometimes a negative net change. For example, acomparison of the number of relationships between the July2007 and January 2008 shows that there was a net change of−1716. Although 47,590 new relationships were added, 49,306other relationships were removed. While the net change is−1716, the total number of changes was actually 96,896. TheComponent History only records the type of changes made toconcepts and descriptions.
2.2. Semantic History mechanism
Currently the Component History focuses mainly on changesin descriptive information within SNOMED CT rather thanchanges in semantic relationships. In this case study, weexamined two types of changes that relate to the seman-tic structure of SNOMED CT rather than descriptive changes.Firstly, we examined the primitive/fully defined status, defin-
ing attributes and long normal form. A change to any of thesethree components may cause a difference in the interpreta-tion of patient records encoded with these concepts. Thesechanges can only be determined by comparing the differ-
Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
444 i n t e r n a t i o n a l j o u r n a l o f m e d i c a l i n f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453
Table 1 – Reason for changes in SNOMED CT between July 2007 and July 2009 releases as recorded in the ComponentHistory.
No Type of change July 2007 January 2008 July 2008 January 2009 Jul 2009
1. Reason not specified 19,868 9919 71,614 13,285 64072. Description status change 5147 4243 56,466 22,074 97003. Description type change 335 12,467 488 291 3294. Description type change, initial
capital status change11 23 6 4 14
5. Description type change, initialcapital status change, languagecode change
0 0 11 0 0
6. Description type change, languagecode change
5 987 400 5 7
7. Fully specified name change 1460 1375 50,901 366 9768. Initial capital status change 22,467 14 29,148 11,280 8809. Initial capital status change,
language code change0 0 2 18 1
10. Language code change 28 1794 501 406 291115
31,
11. Status change 1010
Totals 50,331
ent releases with each other as there is no other trackinginformation available. The primitive/fully defined status canbe determined from the concepts table while the definingattributes are recorded in the relationships table. Secondly,we tracked the position of concepts in the “is a” hierarchy.This includes changes to the top-level hierarchy, the super-type (ancestor) and subtype (descendant) concepts. The “isa” hierarchy plays an important role in subsumption, there-fore any change in the “is a” hierarchy can impact the resultsof subsumption in patient case queries. These changes can
be determined by comparing all the “is a” relationships or byusing a transitive closure table. As the transitive closure tableis currently not part of the standard distribution, we had togenerate it.
Table 2 – SNOMED CT core table entries and type of changes asJanuary 2009.
July 2007 January 2008
Number of records in core tablesConcepts 376,046 378,111Descriptions 1,060,424 1,068,278Relationships 1,359,435 1,357,719Changes to core tablesConcepts
Types of changes to concepts and descriptionsAdded 19,868 9919Status change 6157 5398Minor change 24,306 16,660
Total 50,331 31,977
5 963 9957 3979
977 210,500 58,686 22,584
2.3. SNOMED CT Concept Model mechanism
The SNOMED CT Concept Model defines the semantic relation-ships between concepts, either as part of defining attributesor how post-coordinated expressions should be modeled.The Concept Model is described in the SNOMED CT UserGuide and changes to the Concept Model are recorded in thedocument history section. The Concept Model is also dis-tributed in a machine-readable format called the MachineReadable Concept Model (MRCM). The MRCM however, does
not include a Concept Model mechanism to explicitly iden-tify the changes made according to each bi-annual release ofSNOMED CT.
recorded in the component history between July 2007 and
Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
i n t e r n a t i o n a l j o u r n a l o f m e d i c a l i n f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453 445
Table 3 – Summary of changes to the CORE Problem List between the January and July 2009 releases.
No Type of change Change No change
Total Percent Total Percent
1. Fully specified name 29 0.56% 5153 99.44%2. Preferred term 265 5.11% 4917 94.88%3. Concept status 9 0.17% 5173 99.83%4. Is Primitive 84 1.62% 5098 98.38%5. Long normal form 950 18.33% 4232 81.67%6. Defining attributes 867 16.73% 4315 83.27%
2
TmaiwCJrco(a(
3
ItspfcCcbccswsifpt
3
3Ttficn
7. Supertypes 10908. Subtypes 675
One or more changes 2135
.4. CORE Problem List
he CORE Problem List Subset was designed to facilitate docu-entation and the encoding of clinical records in SNOMED CT
nd is based on dataset submissions from seven healthcarenstitutions [12]. The subset was first released in July, 2009,
hich was based on the January 31, 2009 version of SNOMEDT with 5182 concepts. The subset was updated when the
uly 31, 2009 version of SNOMED CT was released. The cur-ent subset contains 5191 concepts, which includes nine newoncepts that replaced the inactive concepts. The majorityf the concepts are located in the “404684003|Clinical finding
n this case study we examined five major types of changeshrough our four mechanism tracking methods: (1) fullypecified name and preferred term; (2) concept status; (3)rimitive/fully defined status, defining attributes and normalorms; (4) changes in position in hierarchy (top-level hierar-hy, supertypes and subtypes); (5) the effects changes to theoncept Model have on the expressivity of authoring post-oordinated expressions. The comparison we carried out wasased on the first subset release and excluded the nine newoncepts added in the subsequent release since those con-epts were added as a result of changes made to the concepttatus. Overall, 41.2% of the 5182 concepts in the subset under-ent some form of change. A summary of the results are
hown in Table 3. Some changes were minor, such as changesn the ordering of words in the fully specified name and pre-erred term, while others such as the change in concept status,osition of the hierarchy and normal form were more substan-ial and have significant implications.
.1. Fully specified name and preferred term
.1.1. Fully specified namehe changes in the fully specified name were identified using
he Component History table and the change type was classi-ed as a minor. In our case study, we identified four types ofhanges: (1) rephrasing and/or renaming of the fully specifiedame; (2) change in the suffix; (3) adding of diacritics to more
21.03% 4092 78.97%13.03% 4507 86.97%
41.20% 3047 58.80%
accurately describe the concept; (4) correcting of grammaticalerrors.
An example of the renaming of the fully specified namewas when the concept 22325002 was changed from “Gait prob-lem (finding)” to “Abnormal gait (finding)”. A change in thesuffix can indicate a change in the position of the hierar-chy as well. For example, the concept “309158009|Abnormallaboratory findings (finding)|” is now “309158009|Laboratoryfinding abnormal (navigational concept)|”. The semantic tagindicates that it not only changed its position in the hierarchy,but has actually moved to a different top-level hierar-chy concept from “404684003|Clinical finding (finding)|” to“370115009|Special concept (special concept)|”. In the caseof “190818004|Waldenstrom’s macroglobulinemia (disorder)|”,the apostrophe “’s” was removed and a diacritic was added:“190818004|Waldenström macroglobulinemia (disorder)|”. Anexample of correcting a grammatical error is when a space wasadded between the comma and the word “including” for theconcept “218130004|Accident caused by hot liquid and vapor,including steam (event)|”.
3.1.2. Preferred termThe types of changes to preferred terms were similar tothe fully specified name except for the suffix. An exam-ple of a change to a preferred term was the concept“22247000|Dehiscence of surgical wound (disorder)|”, whichwas changed from “Operation wound disruption” in Januaryto “Dehiscence of surgical wound” in July, in accordance to thechanges in the fully specified name. The previous description“Operation wound disruption” was changed from preferredterm to synonym. Descriptions that are no longer consid-ered preferred terms have their description types changed tounspecified or synonym. In addition, the description status isusually changed to retired without a stated reason. Refer toTable 4 for the new description status and type for previouspreferred terms. This table refers to all SNOMED CT conceptsand not just the problem list.
3.2. Concept status
There are two ways to determine if a concept status haschanged. Firstly, a comparison can be made between the cur-
rent concepts table and that of a previous release version.Secondly, a comparison can be made to the Component His-tory table to determine if there is an entry for a component thatincludes a status change. Replacing a duplicate or erroneous
Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
446 i n t e r n a t i o n a l j o u r n a l o f m e d i c a l i n f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453
Table 4 – The resulting changes to the description status and description types of descriptions that are no longerpreferred terms between the January and July 2009 releases.
No Description status Description type July 2007 January 2008 July 2008 January 2009 July 2009
1. Current Unspecified 26 6 0 0 02. Current Synonym 511 336 566 242 2813. Retired without a
concept is straightforward as there is always only one corre-sponding active concept. Out of the retired (7482) and outdated(1345) concepts in SNOMED CT, 2160 (29%) and 1278 (95%) havecorresponding active concepts and all of them are one-to-onereplacements. There are usually at least two or more potentialmeanings for ambiguous concepts. Only 5612 (36%) have exactreplacements, 7249 (47%) have two replacements, 1848 (12%)have three replacements, while the remaining 794 (5%) havebetween four and 21 replacements.
In the problem list subset, there were nine con-cepts that underwent a change in concept status. Fourconcepts are now ambiguous while the other five areconsidered duplicate. In addition, two concepts weremoved to the “363743006|Navigational concept (naviga-tional concept)|” supertype concept and were thereforereplaced by two other concepts. Incidentally, two of theconcepts (“35489007|Depressive disorder (disorder)|” and“309080005|Raised TSH level (finding)|”) that were used toreplace the ambiguous and duplicate concepts were alreadypart of the subset. In the case of “166829003|Serum choles-terol borderline (finding)|”, there are two possible conceptsthat should be used: “442234001|Serum cholesterol borderlinehigh (finding)|” and “442350007|Serum cholesterol borderlinelow (finding)|”.
3.3. IsPrimitive, defining attributes and normal form
Normal forms are created by applying a set of transforma-tion rules set forth in the Transforming Expressions to NormalForms guide [13]. These normalized expressions can be testedfor structural subsumption though a set of pre-defined guide-
Table 5 – Changes to the primitive/fully defined status of conceJuly 2009 releases.
January 2009 Type of change
Primitive
Add newRemove old 1Add new and remove old 1No changes 22
Fullydefined
Add newRemove oldAdd new and remove oldNo changes
Totals 27
0 16 6 81 6 1 0
12,433 2402 899 603
lines. There are a number of factors that may cause the normalform to be altered. They include changes to the IsPrimitivefield, defining attributes and position in the hierarchy. If aprimitive concept is now classified as fully defined, the longnormal form will change because the proximal primitive willno longer refer to itself. If defining attributes are added orremoved, the long normal form will change. If a fully definedconcept has a new primitive supertype inserted between itsposition and the position of the former proximal primitiveconcept, the long normal form will change because the newprimitive supertype concept will be used as the proximal prim-itive concept. A change in the normal form may be the resultof a one or a combination of all three factors. When definingattributes are added or removed, the primitive/fully definedstatus of a concept may remain the same or may change.
3.3.1. IsPrimitiveAs shown in Table 5, in the problem list subset, 2699 (52.1%)concepts remained primitive, 65 (1.3%) primitive conceptswere changed to fully defined, 19 (0.4%) concepts werechanged to primitive, while 2399 (42.3%) concepts remainedfully defined. For example, the concept “74627003|Diabeticcomplication (disorder)|” was considered a primitive in Jan-uary but was changed to fully defined in July. In this case, therewere no changes to the defining attributes but the normal formhas changed because it is no longer its proximal primitive.
3.3.2. Defining attributesA total of 867 (16.7%) changes were made to the definingattributes in the problem list subset, as shown in Table 6. Spe-cial emphasis must be placed on concepts that have defining
pts in the problem list subset between the January and
Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
i n t e r n a t i o n a l j o u r n a l o f m e d i c a l i n f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453 447
Table 6 – The frequency of changes to the defining attributes, supertypes and subtype of concepts in the problem listsubset by the type of change between the January and July 2009 releases.
No Types of change Defining attributes Supertypes Subtypes
Total Percent Total Percent Total Percent
1. No changes 4315 83.27% 4092 78.97% 4507 86.97%2. Both new and old release versions have differences 543 10.48% 212 4.09% 188 3.63%3. Only old release version has more 166 3.20% 378 7.29% 266 5.13%4. Only new release version has more 158 3.05% 500 9.65% 221 4.26%
Totals 5182 100.00% 5182 100.00% 5182 100.00%
Table 7 – Comparison of the defining attributes of “37064009|Hyperproteinemia (disorder)|” between the January 2008and July 2009 releases.
Release/defining attributes January 2008 July 2008 January 2009 July 2009
363714003|Interprets (attribute)| =271934001|Biochemistry test interpretation (observable entity)|
363714003|Interprets (attribute)| = 282294001|Laboratory test finding(observable entity)|
ttributes removed, especially if these concepts were refinedn post-coordinated expression. Refer to Table 7 for the exam-le “37064009|Hyperproteinemia (disorder)|”. Between January008 and January 2009, it included a combination of five defin-ng attributes but all of them were removed in July 2009.
.3.3. Normal formormal forms can be used to test whether two concepts orxpressions are related to each other. Changes to the nor-al forms can significantly impact these relationships. Of the
182 concepts, 950 (18.33%) underwent a change in the normalorm. The change in normal form was the result of a changen the position in the hierarchy, IsPrimitive status, definingttributes, or a combination of all three. Refer to Fig. 1 forhe example “5552004|Disorder of ovary (disorder)|”. In thisase, the long normal form is different because the proximalrimitive was changed from “64572001|Disease (disorder)|” to363124003|Female reproductive system disorder (disorder)|”ecause the concept “363124003|Female reproductive systemisorder (disorder)|” is now a primitive concept.
.4. Change in position in hierarchy
he improvement of logical definitions has resulted inhanges to the inferable subsumption relationships (the “is
January 2009 Release
64572001|Disease (disorder)|:
363698007|Finding site (attribute)|=
15497006|Ovarian structure (body structure)
July 2009 Release
363124003|Female reproductive system disorder (d363698007|Finding site (attribute)|=
15497006|Ovarian structure (body structure)
ig. 1 – Comparison of the long normal forms of “5552004|Disordeleases.
X X X
X
a” hierarchy) [14]. These changes are reflected in the additionof new concepts, retirement of old concepts, the re-organizingof hierarchy and the change in semantic tags. The changes tothe position of a concept in the hierarchy are significant whenit comes to patient case queries.
3.4.1. Top-level hierarchyA comparison between the January and July 2009 releasesshowed that 21 concepts from the “404684003|Clinical find-ing (finding)|” hierarchy were changed, 11 were moved tothe “370115009|Special concept (special concept)|” hierarchywhile the other ten to the “243796009|Situation with explicitcontext (situation)|” hierarchy. In addition two concepts fromthe “71388002|Procedure (procedure)|” hierarchy moved to the“243796009|Situation with explicit context (situation)|” hierar-chy. Refer to Table 8.
Eleven of the 12 concepts that were moved to the“243796009|Situation with explicit context (situation)|”hierarchy are now subtypes of “413350009|Finding withexplicit context (situation)|” or “129125009|Procedure withexplicit context (situation)|” concepts. For example, the
concept “268242003|Congenital subluxation of hip, unilateral(disorder)|” was moved to “268242003|Congenital unilateralsubluxation of hip (situation)|”. In the 11 cases whereby thetop-level hierarchy was changed to “370115009|Special con-
|
isorder)|:
|
er of ovary (disorder)|” between the January and July 2009
cept (special concept)|”, nine are now subtypes of the concept“362955004|Inactive concept (inactive concept)|”. The mainreason is that the concept statuses of these concepts werechanged from current to ambiguous or duplicate. The othertwo concepts are now subtypes of “363743006|Navigationalconcept (navigational concept)|”. There is no history mech-anism to find a replacement concept for navigationalconcepts.
3.4.2. SupertypesThe number of supertypes of a concept can differ greatlybetween two releases of SNOMED CT. Overall, 1090 (21.0%)concepts of the problem list subset underwent a changein supertype concepts in the range of −63 to +14. Refer toTable 6. An example is with the concept “73430006|Sleepapnea (disorder)|”. In the January 2009 release, it had 11 super-type concepts but that was reduced to seven in the July2009 release. Refer to Fig. 2 for the visual representation ofthe supertypes in both releases. The immediate supertypeconcept “1023001|Apnea (finding)|” has been replaced with“39898005|Sleep disorder (disorder)|”.
3.4.3. SubtypesOverall, 675 (13.0%) of the concepts in the subset under-went some form of change, ranging from −403 to +246.Refer to Table 6. A practical example is with the concept“289922002|Mass of ovary (finding)|”. In the January 2009release, the concept did not have any subtype concepts. Thischanged in the July release whereby 112 concepts were addedas subtype concepts. Of these 112 concepts, ten are part of thesubset.
3.5. Example of changes to a single concept
An example whereby a concept went through the most num-ber of changes between the January and July 2009 releaseswas the concept “102594003|Abnormal ECG (finding)|”. Referto Table 9 for a summary of the changes. Firstly, the fullyspecified name was changed from “Abnormal ECG (finding)”to “Electrocardiogram abnormal (finding)” as the acronymECG was spelled out. Secondly, the concept was changedfrom primitive to fully defined as the defining attributes nowfully describe the concept. Thirdly, the defining attributes of“363698007|Finding site (attribute)| = 80891009|Heart struc-
ture (body structure)|” was replaced with “363713009|Hasinterpretation (attribute)| = 263654008|Abnormal (qualifiervalue)|” and “363714003|Interprets (attribute)| = 29303009|Electrocardiographic procedure (procedure)|”. Fourthly, the
0 112 2
5182 5182
long normal form changed as the proximal primitive nolonger referred to itself but to “404684003|Clinical finding(finding)|”. In addition, the long normal form reflected thechanges to the defining attributes. Fifthly, the number ofsupertype concepts was reduced from 13 to five. Three of thesupertype concepts remained the same (“138875005|SNOMEDCT Concept (SNOMED RT + CTV3)|”, “404684003|Clinical find-ing (finding)|” and “102592004|Electrocardiogram finding(finding)|”) while 10 were removed in the July release andtwo new supertypes were added. Sixthly, the number sub-type concepts were reduced from 141 to 136. Six subtypeconcepts were removed while one new subtype conceptwas added, leaving 135 unchanged. Lastly, the preferredterm was changed from “Abnormal ECG” to “Electrocardio-gram abnormal” to reflect the change in the fully specifiedname. The concept status and description status remainedcurrent.
3.6. Concept Model mechanism
There have been changes to the Concept Model in everyrelease since July 2006. We identified three main types ofchanges that have been made to the Concept Model in the lastthree years: (a) change in domain; (b) change in attributes; (c)change in range. Table 10 shows a summary of changes to theConcept Model in the July 2009 release.
3.6.1. Change in domainAn example of a change in a domain was the domain“386053000|Evaluation procedure (procedure)|”, which wasreplaced with the domain “122869004|Measurement pro-cedure (procedure)|”. In this case, the Concept Modelattributes and ranges now apply to the new domain.This change affected 80 concepts that are subtypes of“386053000|Evaluation procedure (procedure)|” but arenot subtypes of “122869004|Measurement procedure(procedure)|”. This change in domain has increased theexpressiveness of the 80 concepts as they are now eli-gible to use six additional Concept Model attributes forpost-coordination.
3.6.2. Change in attributesThere were no changes in the attributes between the Januaryand July 2009 releases. The last time there was a change in an
attribute was in the January 2008 release when the attribute“260669005|Approach (attribute)|” was retired and replacedwith “424876005|Surgical approach (attribute)|”. This newattribute continued to link the domain “71388002|Procedure
Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
i n t e r n a t i o n a l j o u r n a l o f m e d i c a l i n f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453 449
F |Slee(
((
3Tisars
ig. 2 – Comparison of the supertype concepts of “73430006right) 2009 releases.
procedure)|” to the range “103379005|Procedural approachqualifier value)|”.
.6.3. Changes in rangehe July 2009 release saw a variety of changes in range,
ncluding the replacement of multiple ranges with a single
upertype range, the addition of a range, and the removal ofrange. The replacement of a set of ranges with a supertype
ange occurred twice. The first time was when the anatomicalites of “91723000|Anatomical structure (body structure)|” and
p apnea (disorder)|’ between the January (left) and July
“280115004|Acquired body structure (body structure)|” werereplaced with “442083009|Anatomical or acquired body struc-ture (body structure)|”. The domains that were affected were“404684003|Clinical finding (finding)|”, “71388002|Procedure(procedure)|” and “123038009|Specimen (specimen)|”. Theeffect was minor as overall, the range increased only by the
single supertype concept as there were no additional sub-type concepts that were not previously specified as part ofthe range. The problem list subset has 4529 concepts in the“404684003|Clinical finding (finding)|” domain and 412 con-
Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
450 i n t e r n a t i o n a l j o u r n a l o f m e d i c a l i n f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453
Table 9 – Summary of changes for the concept “102594003|Abnormal ECG (finding)|” between the January and July 2009releases.
Characteristic January 2009 July 2009
Fully specified name Abnormal ECG (finding) Electrocardiogram abnormal (finding)Concept status Current CurrentPreferred term Abnormal ECG Electrocardiogram abnormalDescription status Current CurrentLong normal form 102594003|Abnormal ECG (finding)|:
363698007|Finding site (attribute)| =80891009|Heart structure (body structure)|
• 301095005|Cardiac finding (finding)|• 250906000|Cardiac investigative finding (finding)|• 106063007|Cardiovascular finding (finding)|• 118234003|Finding by site (finding)|• 301857004|Finding of body region (finding)|• 298705000|Finding of region of thorax (finding)|• 302292003|Finding of trunk structure (finding)|• 301296002|Mediastinal finding (finding)|• 277775005|Test finding (finding)|• 406123005|Viscus structure finding (finding)|
• 442618008|Abnormal finding on evaluationprocedure (finding)|• 441742003|Evaluation finding (finding)|
Subtypes (onlychanges shown)
• 251136001|Borderline abnormal ECG (finding)|• 59462000|Decreased EKG voltage (finding)|• 251146004|Low QRS voltages (finding)|• 251147008|Low QRS voltages in the limb leads(finding)|
cord
KG (fi
• 20329001|Low ventricular voltage by EKG (finding)|
• 251148003|Low QRS voltages in the preleads (finding)|• 20329001|Low ventricular voltage by E
cepts in the “71388002|Procedure (procedure)|” domain. Thesecond time it occurred was when four subtype concepts wereremoved from the range for the attribute “272741003|Laterality(attribute)|”. They were removed as “82353008|Side (qualifiervalue)|”, a supertype concept, was already specified and thefour concepts were redundant. In this case, the concepts inthe range remained the same. The concepts that were affectedwere the 46 concepts that are from the “123037004|Body struc-ture (body structure)|” domain.
An example of the addition of a range was when“441862004|Infectious process (qualifier value)|” was addedto the domain “71388002|Procedure (procedure)|” throughthe Concept Model attribute “370135005|Pathological process(attribute)|”. In this case, the range contained just two con-cepts. The addition of 441862004|Infectious process (qualifiervalue)|” as a range enabled 6900 concepts to be redefined withthe sanctioned qualifier values. The domain that was affectedwas also “404684003|Clinical finding (finding)|”. An exam-ple of the removal of a range was when “281296001|Resultcomments (qualifier value)|” was removed from the domain“404684003|Clinical finding (finding)|” through the Concept
Model attribute “363713009|Has interpretation (attribute)|”.This reduced access to 20 qualifier value concepts andany post-coordinated expressions that used these qualifiervalues are using unsanctioned concepts. The removal of
ial
nding)|
“281296001|Result comments (qualifier value)|” as a rangecaused 104 concepts to have their defining attributesremoved. The domain that was affected was again the“404684003|Clinical finding (finding)|”.
4. Discussion
As demonstrated in our results, there are many changes thatSNOMED CT is undergoing. These differences can be theresult of changes in medical knowledge which find their wayinto SNOMED CT or due to changes in SNOMED CT itself,in which the (unchanged) knowledge is more adequatelyrepresented. An example of the former is when the concept“442696006|Influenza due to Influenza A virus subtype H1N1(disorder)|” was added in July 2009 while an example of the lat-ter is when the concept “371622005|Elevated blood-pressurereading without diagnosis of hypertension (finding)|” wasmore clearly defined as a situation as opposed to a finding inJuly 2009.
As part of the SNOMED CT documentation, the IHTSDO
publishes changes that pertain to the maintenance of thecontent, such as the remodeling of top level categories andeditorial decisions (e.g., the redefining of active concepts asjust current concepts only and not limited status). How-
ver, the rationale of changes at the individual conceptevel is not published. Between the January and July 2009eleases, 41.2% (2135) of the concepts in the problem listubset underwent changes in the form of fully specifiedame/preferred term, concept status, primitive/fully definedtatus, defining attributes, normal forms, top-level hierar-hy, supertypes and/or subtypes. In addition, changes to theoncept Model can affect the expressivity of creating post-oordinated expressions. While it can be assumed that thesehanges are made for improving the quality of SNOMED CT,t would be helpful to publish the motivation for the changes
ade to the terminology using a framework such as the oneescribed by Bakhshi-Raiez et al. [15].
Organizations that have implemented or plan on imple-enting SNOMED CT must be aware of the types of version
hanges that can occur and examine each SNOMED CT releaseo ensure there are no unexpected changes that may causeheir information systems to produce unexpected results.s some of these changes are not well published, one mustay attention to the subtle changes in each release in ordero reconcile them with their existing system. Fortunately,hese changes can be tracked easily without the need ofophisticated algorithms or methods. Therefore, beforeeploying a new release of SNOMED CT, organizations shouldetermine the implications of these changes on their existingerminology set.
.1. Implication of findings
.1.1. Inactive conceptsnactive concepts should not be used in patient records. There-ore one will have to determine which inactive concepts within
r value)|(qualifier value)|alifier value)|
their existing terminology set can be replaced by active con-cepts and log an audit trail of these changes. Selecting one ofthe concepts as a replacement for all historical patients maynot be viable as each case needs to be evaluated. This is proba-bly not feasible as significant resources will need to be devotedto the review process and even then a conclusion may not bereached as to which is more suitable. The effects of the inacti-vation of a qualifier have even further reaching consequences.Qualifiers are used in post-coordinated expressions andtherefore all expressions that include the newly inactivatedqualifier value needs to be reviewed. The previous example of“166829003|Serum cholesterol borderline (finding)|” highlightsthis issue. Since the concept status of “75189007|Borderline(qualifier value)|” was changed to ambiguous in the Julyrelease, it may now refer to “442779003|Borderline low (qual-ifier value)|” or “442777001|Borderline high (qualifier value)|”.Patient records that have been encoded with “borderline” maybe made up of a combination of “highs” and “lows” and there-fore it is not possible to automatically update all expressionswith an active concept. Depending on the number of recordsencoded with this qualifier, it may not be feasible to retrieveeach record and analyze the context in which “borderline” wasused. Since ambiguous concepts usually have more than onepotential meaning, organizations may have to come to therealization that their electronic records will contain a com-bination of active and inactive SNOMED concepts.
When inactive concepts are moved to the“370115009|Special concept (special concept)|” hierarchy,
this will impact the testing of equivalency and subsumptionin expressions. If there are inactive concepts in the patientrecords, one will need to query the historical relationshipsto determine how these inactive concepts used to fit into
Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
i c a l
452 i n t e r n a t i o n a l j o u r n a l o f m e d
the hierarchy. This may yield incomplete results as notall corresponding active concepts are one-to-one matches.Pre-defined queries that “roll up” to a certain concept may nolonger include the same concepts as subtype concepts. Thismay lead to different frequency counts based on previouslydefined queries. Existing reports that show the trend ofcertain diseases may now be affected as now a different setof concepts are used.
In addition to keeping track of inactive concepts, conceptsthat are moved to subtypes of “363743006|Navigational con-cept (navigational concept)|” should also be replaced eventhough they still main an active concept status since navi-gational concepts are used to support navigation and shouldnot be used in patient records or aggregating information.
4.1.2. Changes to descriptionsChanges in the preferred term are of no consequence to testingfor equivalency and subsumption. They may, however, causesome difficulties in encoding if clinicians who are familiarwith the previous preferred terms cannot locate their replace-ments easily. If a software application has been set up toautomatically display a concept only using the preferred term,clinicians should be made aware of the changes. In Section 3.1,the example was given when the preferred term “Operationwound disruption” was replaced with “Dehiscence of surgicalwound.” The adding of diacritics can reduce the effectivenessof lexical matching and therefore data entry if an applicationdoes not allow for diacritics input or is unable to normalizediacritics.
4.1.3. Changes to the Concept ModelChanges to the Concept Model can have an impact on thevalidity of post-coordinated expressions. The addition of arange can help to increase the expressiveness when creat-ing post-coordinated expressions. It can also create accesspoints to qualifier values that previously could not be usedthrough the Concept Model. When a range is removed from anattribute, it can cause conformance issues as post-coordinatedexpressions that used those ranges no longer adhere to thecurrent version of the Concept Model. The removal of a rangealso decreases the expressiveness of post-coordinated expres-sions and can reduce access to qualifier values. Thereforepost-coordinated expressions need to be reviewed to ensurethey conform to the Concept Model. Since the problem list sub-set consists of only pre-coordinated concepts, we were unableto ascertain the effects of the changes to the Concept Modelon post-coordinated expression. Wade et al. reported that 71%of their post-coordinated expressions required updating [6].However, in the last two releases of SNOMED CT, the Con-cept Model has not undergone as significant changes as it didbetween the January 2005 and July 2006 releases.
4.1.4. Patient case queriesIn our analysis, we found that a concept can change betweenprimitive and fully defined and vice versa for a number ofreasons including the adding, removal as well as adding and
removal of defining attributes. In addition, even if there is nochange in the defining attributes, concepts can still changebetween primitive and fully defined and vice versa. Thereforeeven if the IsPrimitive status does not change, it is still impor-
i n f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453
tant to check if any changes have been made to the definingattributes and the long normal forms. Since the long normalform is used in structural subsumption to test for equivalencyand subsumption, it is important to track these changes.
The changes to the “is a” relationships can have an impacton patient case queries when testing for equivalency andsubsumption. By changing the position in the hierarchy,concepts may now contain different supertype and subtypeconcepts. Concepts such as “46635009|Diabetes mellitus type1 (disorder)|” and “399923009|Rheumatoid arteritis (disorder)|”that were considered “421976005|Hypersensitivity disorder(disorder)|” are now only considered “75934005|Metabolicdisease (disorder)|” and “128139000|Inflammatory disorder(disorder)|” respectively. Another example is when con-cepts are moved to a different hierarchy. For example,querying for all subtypes of “69031006|Excision of breasttissue (procedure)|” would have yielded concepts such as“8115005|Unilateral radical mastectomy (procedure)|” in Jan-uary 2009 but now that the concept was moved to the“243796009|Situation with explicit context (situation)|” hier-archy and is now known as “8115005|Unilateral radicalmastectomy (situation)|”, patient records encoded with thatconcept would not have been retrieved.
These changes can cause inconsistencies in reportingtrends as the increases and decreases are due to the re-structuring of the “is a” relationships as opposed to anychanges in actual patient cases. If there are pre-definedqueries and the number of subtype concepts keeps chang-ing, it will cause inaccurate reporting over different versionsof SNOMED CT. It is also important to know the subtype con-cepts of each concept so that they can be tested to determine ifthey are being double counted such as diabetes mellitus type1 is now considered a metabolic disease and inflammatorydisorder but not a hypersensitivity disorder.
4.2. Future work
As the IHTSDO has now introduced a new format for distribut-ing SNOMED CT, called release format 2 (RF2), organizationscan take advantage of the ability to re-construct previousSNOMED CT releases and to compare and track the changesmade over time. Previously organizations would have had toobtain previous releases to make the comparisons. Futurework would include analyzing the RF2 and to compare theresults of auditing methods to determine if the quality ofSNOMED CT has improved over time.
4.3. Limitations
A limitation of this study is that it did not employ sophisti-cated auditing algorithms and methods such as those in usedin previous studies [13]. However, these changes can be stillbe tracked without the use of auditing methods as the objec-tive was to track the changes and discuss the implications asopposed to auditing the content and accuracy of the changes.In addition, the changes were tracked on a comparatively
small subset of SNOMED CT, with just under 5200 conceptswhile entire subset of SNOMED CT contains over 388,000 con-cepts. The changes may not be reflective of the other conceptsin SNOMED CT.
Journal Identification = IJB Article Identification = 2729 Date: April 21, 2011 Time: 11:3 am
i n t e r n a t i o n a l j o u r n a l o f m e d i c a l i n
Summary pointsWhat was already known on the topic?
• SNOMED CT is changing at a rapid rate with thousandsof concepts being added and retired with each release.
• Changes in the Concept Model can cause previouslyencoded post-coordinated expressions to no longerconform to the Concept Model.
What this study added to our knowledge?
• The repositioning of concepts in the “is a” hierar-chy and changes to concept definitions can causeunexpected results when testing for equivalency andsubsumption and this can lead to inconsistencies inaggregate reports.
• The meaning of a concept can change constantlythrough each release of SNOMED CT as defining
5
IstctCii
A
DDpqrd
C
DcCHwe
r
[15] F. Bakhshi-Raiez, R. Cornet, N.F. de Keizer, Development andapplication of a framework for maintenance of medicalterminological systems, Journal of the American Medical
attributes are added and removed.
. Conclusion
n this paper, we have identified four types of changes (fullypecified name and preferred term; concept status; primi-ive/fully defined status, defining attributes and normal forms;hanges in position in hierarchy (top-level hierarchy, super-ypes and subtypes) that occur over time as new SNOMEDT releases are introduced. Keeping track of these changes is
mportant as they are not well published and have an impactn patient case queries and the accuracy of patient records.
uthor contributions
ennis Lee conducted the analysis and wrote the initial draft.r Ronald Cornet and Dr Francis Lau reviewed the results,rovided suggestions, critical feedback and revised subse-uent drafts. All three authors reviewed the comments by theeviewers, addressed the issues raised and approved the finalraft for submission.
onflict of interest statement
r. Ronald Cornet is the co-chair of the Implementation Spe-ial Interest Group and a member of the Quality Assurance
ommittee and of the Member Forum at the Internationalealth Terminology Standards Development Organization,hich publishes SNOMED CT. His position at the IHTSDO, how-
ver, had no bearing on the research study or results.
f o r m a t i c s 8 0 ( 2 0 1 1 ) 442–453 453
e f e r e n c e s
[1] Y. Wang, M. Halper, H. Min, Y. Perl, Y. Chen, K.A. Spackman,Structural methodologies for auditing SNOMED, Journal ofBiomedical Informatics 40 (2007) 561–581.
[2] G. Jiang, C.G. Chute, Auditing the semantic completeness ofSNOMED CT using formal concept analysis, Journal of theAmerican Medical Informatics Association 16 (2009)89–102.
[3] W. Ceuster, B. Smith, A. Kumar, C. Dhaen, Ontology-basederror detection in SNOMED CT, Studies in Health Technologyand Informatics 107 (Pt. 1) (2004) 482–486.
[4] R. Cornet, A. Abu-Hanna, Auditing description-logic-basedmedical terminological systems by detecting equivalentconcept definitions, International Journal of MedicalInformatics 77 (May (5)) (2008) 336–345 (Epub 2007 August10).
[5] S.K. Nachimuthu, L.M. Lau, Practical issues in usingSNOMED CT as a reference terminology, Studies inHealth Technology and Informatics 129 (Pt. 1) (2007)640–644.
[6] G. Wade, S.T. Rosembloom, The impact of SNOMED CTrevisions on a mapped interface terminology: terminologydevelopment and implementation issues, Journal ofBiomedical Informatics 42 (2009) 490–493.
[7] J. Ingenerf, T. Beisiegel, A version management system forSNOMED CT, Studies in Health Technology and Informatics136 (2008) 827–832.
[8] International Health Terminology Standards DevelopmentOrganization, Compositional Grammar for SNOMEDCT Expressions in HL7 Version 3, December 23, 2008,p. 14.
[9] International Health Terminology Standards DevelopmentOrganization, SNOMED CT User Guide, July 2009International Release, p. 57.
[10] International Health Terminology Standards DevelopmentOrganization, SNOMED CT Technical Implementation Guide,July 2009 International Release, p. 54.
[11] International Health Terminology Standards DevelopmentOrganization, SNOMED CT Technical Reference Guide, July2009 International Release, p. 36.
[12] The CORE Problem List Subset of SNOMED CT.<http://www.nlm.nih.gov/research/umls/Snomed/core subset.html> (accessed 12.01.10).
[13] International Health Terminology Standards DevelopmentOrganization, SNOMED CT Transforming Expressions toNormal Forms, January 31, 2008, p. 10.
[14] K.A. Spackman, Rates of change in a large clinicalterminology: three years experience with SNOMED clinicalterms, American Medical Informatics Association AnnuSymp Proc (2005) 714–718.
Informatics Association 15 (September–October (5)) (2008)687–700 (Epub 2008 June 25).
