BMC Medical Informatics and

Decision Making BioMed Central

Proceedings Open Access

Experiences mapping a legacy interface terminology to SNOMED
CT
Geraldine Wade*1 and S Trent Rosenbloom2

Address: 1Information Management, Hewlett-Packard Company, Atlanta, GA, USA and 2Department of Biomedical Informatics, Vanderbilt
University, Nashville, TN, USA
Email: Geraldine Wade* - [email protected]; S Trent Rosenbloom - [email protected]
* Corresponding author

from First European Conference on SNOMED CT

Copenhagen, Denmark. 1–3 October 2006

Published: 27 October 2008

BMC Medical Informatics and Decision Making 2008, 8(Suppl 1):S3 doi:10.1186/1472-6947-8-S1-S3

<supplement> <title> <p>Selected contributions to the First European Conference on SNOMED CT</p> </title> <editor>Stefan Schulz and Gunnar O Klein</editor> <sponsor> <note>Publication of this supplement was supported by EU Network of Excellence "Semantic Interoperability and Data Mining in Biomedicine"</note> </sponsor> <note>Proceedings</note> <url>http://www.biomedcentral.com/content/pdf/1471-6947-8-S1-info.pdf</url> </supplement>

This article is available from: http://www.biomedcentral.com/1472-6947/8/S1/S3

© 2008 Wade and Rosenbloom; licensee BioMed Central Ltd.
This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0),
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract
Background: SNOMED CT is being increasingly adopted as the standard clinical terminology for
health care applications. Existing clinical applications that use legacy interface terminology need to
migrate to the preferred SNOMED CT standard. In this paper, we describe our experience and
methodology for mapping concepts from a legacy system to SNOMED CT.
Methods: Our approach includes the establishment of mapping rules between terminologists and
back and forth collaboration of the mapped results through one or more iterations in order to
reach consensus on the final maps.
Results: We highlight our results not only in terms of the number of matches, quality of maps, use
of post-coordination, and multiple maps but also include our observations about SNOMED CT
including inconsistencies, redundancies and omissions related to our legacy mapping.
Conclusion: Our methodology and lessons learned from this mapping exercise may be helpful to
other terminologists who may be similarly challenged to migrate their legacy terminology to
SNOMED CT. This mapping process and resulting discoveries about SNOMED CT may further
contribute to refinement of this dynamic, clinical terminology standard.

Background and often used in stand-alone applications that have

Institutions and Electronic Health Record (EHR) system become outdated or need modification for clinical utility.
vendors are being increasingly challenged to use recog- Using a standard terminology rather than a legacy inter-
nized standard terminologies, such as SNOMED CT [1]. face terminology may help make EHR systems be interop-
Using standard terminologies often requires that system erable with other such systems, drive decision support
developers migrate away from legacy interface terminolo- algorithms, enable data aggregation for quality analysis/
gies [2] by mapping them to a reference terminology. outcomes measurements, among other tasks [3-5]. Using
Interface terms have largely been proprietary in nature standardized terminology within applications may sup-

Page 1 of 6
(page number not for citation purposes)
BMC Medical Informatics and Decision Making 2008, 8(Suppl 1):S3 http://www.biomedcentral.com/1472-6947/8/S1/S3

port evidence-based initiatives, improve patient safety as (e.g. Leukemia, Ulcerative Colitis, Sinusitis, Breast Cancer),
well as meet new regulatory requirements [6] as standards time (e.g. Date of Last Menstrual Period), objects (e.g. Shunt
are adopted both nationally and internationally. Vander- for Hemodialysis Access, Implanted Cardiac Device), proce-
bilt University Medical Center (VUMC) in Nashville, TN, dures (e.g. Appendectomy, Venous Access Device Placement),
USA, has developed a clinical interface terminology for scales (e.g. Patient Pain Scale, Epworth Sleep Scale Score)
use in its EHR system components, including a structured and social (e.g. Unemployed, Family Makeup). Several con-
entry tool designed to support clinical documentation. cepts did not appear to fit any particular category or were
The terminology was designed as an outgrowth of the one less well-defined (e.g. Has a Gun in the House, Wears a Hel-
created in the 1980s to support the Internist/QMR diag- met while Riding a Motorcycle).
nostic expert and decision support system [7]. The inter-
face terminology includes concepts for general medical Before mapping, there was general agreement between the
evaluation, including those covering history, exam and two terminologists on mapping rules including how the
diagnoses. quality of the mapping relationships would be defined
(see below) and how post coordinated concepts would be
Methods represented.. For example, several source concepts were
The terms representing legacy interface concepts were entities that were "auscultated" (e.g. Heart Murmur Auscul-
extracted from the Vanderbilt EHR systems in a flat file tated, Abdominal Bruit Auscultated). These were all to be
format (i.e. Excel spreadsheet) for evaluation and map- mapped similarly using agreed upon post-coordinated
ping. Concepts and their unique identifiers were obtained concept groupings in SNOMED CT (e.g. Finding by aus-
(e.g. ID02964: Anaphylactic Shock) sequenced by a pro- cultation [finding] Associated with [attribute]).
gressive list of concept identifier numbers. No corre-
sponding clinical context from the computer programs The concept mapping process involved 4 steps. The first
using the terminology was initially provided. The con- step was to group the legacy (source) concepts into rele-
cepts related to history (e.g. Ethanol Dependence History), vant clinical categories (Table 1). Concepts that included
history or symptom (e.g. Myalgia History or Symptom), terms such as Auscultated or Palpated were grouped simi-
physical examination (e.g. Heart Sound S3 Auscultated, Ear larly and were assessed as being part of a physical exami-
Erythema Observed, Tactile Fremitus Palpated), diagnoses nation.

Table 1: Grouping of legacy interface terminology concepts with corresponding examples

CATEGORIES LEGACY CONCEPTS

(# of concepts)

Historical (500)
Allergy (12) ALLERGY TO LATEX
Family History (40) FAMILY HISTORY OF NEUROPATHY
History (80) DYSLEXIA HISTORY
History or Symptom (353) DIAPHORESIS HISTORY OR SYMPTOM
Ob-Gyn History (8) NUMBER OF CHILDREN
Risk Factors (3) CARDIAC RISK FACTORS
Physical Exam (972)
Auscultation (68) VENOUS HUM AUSCULTATED
Elicited (200) PULSUS PARADOXUS ELICITED
Measurement (56) BODY MASS INDEX QUANTITATIVE MEASURED
Observation (487) AGITATION OBSERVED
Palpation (148) HEART THRILL PALPATED
Percussion (13) LIVER SPAN QUANTITATIVE PERCUSSED
Other (530)
Activities and Functions (49) USE OF AMBULATION ASSISTIVE DEVICES
Chief Complaint (12) CHIEF COMPLAINT EXPOSURE TO CHEMICAL
Clinical finding (299) MENINGITIS
Date (56) DATE OF FIRST POLIO VACCINATION
Devices (5) INDWELLING URINARY CATHETER
Misc. (7) PATIENT TRANSFERRED FROM
Personal and Social (27) TYPE OF LIVING ACCOMODATION
Procedure (44) REPAIR OF TETRALOGY OF FALLOT
Referral (5) REFERRAL FOR ABNORMAL ECHOCARDIOGRAM
Scales and scores (26) EPWORTH SLEEP SCALE SCORE

Page 2 of 6
(page number not for citation purposes)
BMC Medical Informatics and Decision Making 2008, 8(Suppl 1):S3 http://www.biomedcentral.com/1472-6947/8/S1/S3

Those concepts that included History, History or Symptom, selected as targets. The target concept used in the result set
Family History, Risk Factors, etc. were also grouped simi- included the fully specified name designated by SNOMED
larly and were assessed as being historical. Some concepts CT. As each map was recorded, a separate entry was also
were grouped based on the terminologists' judgment that recorded as to the quality of the relationship between and
included underlying clinical knowledge/domain exper- source legacy interface terms and target SNOMED CT con-
tise. For example, concepts such as Supports Self on Fore- cepts. A source concept that mapped to a semantically
arms While Prone and Plays "Pat-A-Cake" Responsively were equal single SNOMED CT concept was qualified as equal.
known to be observations of one's development status
and Inguinal Herniorrophy and Mastectomy were known An equal qualifier was also given to maps that used com-
surgical procedures. A concept such as Taking Anticoagu- bined target concepts using the post coordination guide-
lant Medication could have been placed in more than one lines developed by the SNOMED CT Concept Model
grouping (e.g. History or Activities and Functions) but a Working Group [11], the SNOMED CT Users guide [12]
single group (i.e. Activities and Functions) was subjec- and the Technical Implementation guides [13].
tively selected for mapping purposes. Some of the con-
cepts were categorized as miscellaneous when they did They were noted under a separate category (see Results,
not appear to be part of logical group (e.g. Patient Trans- Table 2). The same was done for relationships that were
ferred From, Follow Up Evaluation For, etc). By grouping the qualified as related but not equal to a single target concept
concepts in this way, most could be correlated with the or targets. A source concept that was not mappable to tar-
upper level SNOMED CT categories/axes. Additionally, get concepts in SNOMED CT was recorded as "No Match".
groups of similar concepts could be mapped in a consist- Some final maps included IS A relationships since the
ent way using similar rules. This was most important for source concept only appeared to relate to higher-level
representing SNOMED CT concepts requiring post-coor- concepts in SNOMED CT.
dination.
The fourth step was to share the resulting groups of maps
The second step involved searching the SNOMED CT with the second terminologist for validation and com-
knowledgebase (January 2005) [8] for concepts within mentary. Each concept map was agreed to or was com-
each of the groupings. Both proprietary search tools [9] mented upon for further review/discussion. The maps
and the Clue Browser [10] were used. Concepts were were then returned to the first terminologist. Comments
searched for and selected by using their word matching included requests for remapping, additional clarification
and/or synonym matching with consideration of where as to why a given target was chosen and clinical explana-
they fit within in a given hierarchy. If the source concept tions as to why the SNOMED synonym was incorrect or
was a procedure, a corresponding target concept in the inconsistent. On occasion, additional context was pro-
SNOMED CT procedure axis was selected. vided to the first terminologist based on knowledge of the
actual clinical context. For example, the concept Ortho-
The third step was to record the selected target concepts in pedic Surgery could be interpreted as referring to the
a spreadsheet adjacent to the source (legacy) concept. Orthopedic Surgery Department or to an Orthopedic sur-
Only active non-limited SNOMED CT concepts were gical procedure.

Table 2: Results of legacy concept mapping to SNOMED CT with examples

Number of concepts Mapped relationship(s) Source concept(legacy) Target concept (SNOMED CT)
(2002 = total)

1510 302 mapped to equal single target OSTEOPOROSIS Osteoporosis (disorder)

(75%) concept
1208 mapped to equal post-coordinated POST-ICTAL PSYCHOSIS Psychotic disorder (disorder)
targets Associated with (attribute)
Post-ictal state (finding)
396 34 mapped to related single target BEHAVIORAL DISTURBANCE Problem behavior (finding)
(20%) concept
362 mapped to related post-coordinated PELVIS MUSCULAR TONE FLACCID Finding by palpation (finding)
targets PALPATED Associated with (attribute)
Poor pelvic muscle tone (finding)
70 Mapped to parent target concept only (IS SMILES TO IMAGE OF PARENTS FACE Child developmental finding (finding)
(4%) A)
26 No map HEART SOUND CLICK
(1%) AUSCULTATED

Page 3 of 6
(page number not for citation purposes)
BMC Medical Informatics and Decision Making 2008, 8(Suppl 1):S3 http://www.biomedcentral.com/1472-6947/8/S1/S3

The process of back and forth collaboration between the Discussion

two terminologists (GW, STR) continued for two or three SNOMED CT is a dynamic, scientifically validated clinical
iterations until all maps were completed. health care terminology and infrastructure [14] that is
being increasingly adopted as the preferred terminology
Results for the representation of clinical information. As health-
2002 legacy interface terms from VUMC were evaluated. care providers, payers and government officials focus on
Among the resulting final maps to SNOMED CT (Table developing interoperable electronic health networks, data
2), there were 1510 concepts that were rated by two termi- standards including SNOMED CT are being increasingly
nologists (GW, STR) as having semantically equivalent incorporated into new and existing healthcare applica-
matches. In this group, 302 legacy concepts mapped each tions to meet data sharing needs. Transforming legacy and
to single SNOMED CT concepts and 1208 legacy concepts proprietary terminologies into standards will be required
mapped to a combination of post-coordinated concepts. for clinical utility. Such legacy interface terminologies,
Maps that were related but not semantically equal like the one we have described, may consist of an aggre-
included 34 single concept maps and 362 post-coordi- gate of single concepts or concept phrases and not part of
nated maps. Seventy concepts were designated as having a structured, controlled terminology. Thus alignment
an IS A relationship as they appeared to represent an methods that have been described previously [15-17]
appropriate child concept relative to a SNOMED CT con- using algorithms to compare structured knowledge
cept. Twenty-six concepts were not matched (e.g Heart sources could not be used. Formal definition description
Sound Click Auscultated, Presyncope, Low Pitched Bowel logics (DLs) have also been shown to aid in mapping
Sounds Auscultated,). Among the post-coordinated maps, between terminologies by providing concept and role def-
580 were more complex in that several attribute-value initions with explicit semantics [18]. These, too, were
pairs were used (e.g., Precordial Cardiac Impulse Intensity absent from the legacy interface terminology in this eval-
Palpated mapped to Finding by palpation (finding) + uation. To offset some of these limitations, we felt that it
Associated with (attribute) + Finding of pulse volume was important to group concepts into clinically relevant
(finding) + Interprets (attribute) + Precordial pulsation, categories ahead of the actual mapping in order to provide
function (observable entity)). Additional results showed some consistency for mapping of concepts within a given
that 9 of the legacy concepts mapped to more than one group. Lexical associations (i.e. auscultated, palpated, his-
SNOMED CT concept (i.e. 1 to many relationship). In tory or symptom, etc) included in many of the concept
some instances, there were maps to two equivalent strings helped guide some of the obvious groupings. The
SNOMED CT concepts (i.e. 1 to 1 relationship) Fig. 1 terminologists could then discuss and agree prospectively
on mapping rules that would apply generally as well as to
In addition, some of the SNOMED CT target concepts the differing groups of concepts
seemed to be formatted inconsistently (e.g. Left popliteal
artery structure (body structure) and Structure of right After establishing agreed-upon mapping rules, there were
popliteal artery (body structure)). All of the final outputs a series of process steps involving searching, recording
were recorded in a flat file/Excel spreadsheet and given to and qualifying relationships among the mapped con-
the IT group for future consideration/integration into the cepts. There was ongoing collaboration – validation, dis-
current clinical application. cussion and commentary for each group of maps. This
was critical to achieving eventual consensus on the final
maps.

One to m
One to many relationship
EBSTEIN ANOMALY Ebstein's anomaly of common atrioventricular valve (disorder)
Ebstein's anomaly of tricuspid valve (disorder)

One to one relationship

Disorder due to work-related activity accident (finding)
INJURY RELATED TO WORK
Accident while engaged in work-related activity (finding)

Figure 1examples showing different relationships to target maps

Mapping
Mapping examples showing different relationships to target maps.

Page 4 of 6
(page number not for citation purposes)
BMC Medical Informatics and Decision Making 2008, 8(Suppl 1):S3 http://www.biomedcentral.com/1472-6947/8/S1/S3

In our experience, it is critical to have terminologists with application, may give further insight as to how it may be
considerable clinical background or domain expertise best to proceed with integration. For instance, a more fre-
who could apply their knowledge to the grouping and quently used clinical concept such as "myocardial infarc-
mapping of concepts whose meaning may not be obvious tion" is well represented in SNOMED CT [21] and could
by the description alone. In this evaluation of legacy inter- be immediately deployed for use within an application. A
face concepts, no corresponding clinical context was given less frequently used concept, such as "Epworth sleep scale
ahead of the first mapping iteration and this led to some score" is not currently represented in SNOMED CT but
initial errors. Perhaps by providing some clinical context may not be critical data for capture as a "standard" as it
with a list of legacy concepts there would be better seman- would be much less likely to be used in decision-support
tic maps with SNOMED CT. By grouping legacy concepts algorithms or patient safety measures.
into similar categories prospectively and by using map-
ping rules in a consistent manner to each group, future Conclusion
changes made to SNOMED CT may be more readily Using these 2002 concepts as a typical example of what
applied to your mapped legacy terminology (e.g. If new other terminologists may face when challenged with tran-
attribute-value pairs are added or previous guidelines sitioning their proprietary concepts to standardized termi-
revised, new pairs of concepts can be consistently nology, this methodology can be applied using a
applied.) systematic approach – starting with legacy concept group-
ing and establishment of rules for mapping concepts that
We observed that this process exposed not only differ- are grouped similarly as well as establishing consensus
ences between the two terminologists in their semantic (between terminologists) for how rules will be applied
interpretation of concepts but also highlighted areas in and for how Attribute-Value pairs will be applied to par-
SNOMED CT that were redundant, inadequate or defi- ticular groups of concepts. Such mapping and analysis
cient. For example, we did not think that "depression contributes to the improvements in SNOMED CT as clin-
(finding)" and "sadness" were semantically equal as ical concepts are continuously added and modified
defined by SNOMED CT. We found that "rectocele" was (through submissions and inquiries).
used as a synonym for the preferred display concept of
"female proctocele without uterine prolapse (disorder)", Competing interests
even though there are rare instances when it occurs in a The authors declare that they have no competing interests.
male. This example also highlighted the discovery that
some of the preferred display concepts led to a change in Authors' contributions
a map upon review. Even though there may have been an Both authors (GW and STR) designed the study, per-
exact match to a synonym in SNOMED CT, the preferred formed and evaluated the concept mappings, drafted and
display concept, on occasion, suggested an alternate approved the final manuscript.
meaning that led to a re-examination of the map. This
mapping exercise also led to the identification of concepts Acknowledgements
that needed to be added to SNOMED CT. Despite these The project was supported in part by a grant from the United States
deficiencies and omissions, there was overall good clinical National Library of Medicine (Rosenbloom, 2K22 LM008576-02).
concept representation of this legacy interface terminol-
This article has been published as part of BMC Medical Informatics and Deci-
ogy set in SNOMED CT. Also, it is useful to note that
sion Making Volume 8 Supplement 1, 2008: Selected contributions to the
SNOMED CT is dynamic – a work in progress – with bian- First European Conference on SNOMED CT. The full contents of the sup-
nual updates and new releases. As a standards organiza- plement are available online at http://www.biomedcentral.com/1472-6947/
tion, it is open to participation and invites submissions 8?issue=S1.
for additions and modifications. SNOMED CT editors rely
on inputs from users. This makes it most suitable for the References
complexities of clinical medicine. Efforts to extend termi- 1. National Committee on Vital and Health Statistics: Report to the Sec-
retary of the U.S. Department of Health and Human Services Uniform Data
nologies such as SNOMED CT into ontologies offer addi- Standards for Patient Medical Record Information . July 6, 2000
tional sources of discriminating reviews [19,20]. 2. Rosenbloom ST, Miller RA, Johnson KB, Elkin PL, Brown SH: Inter-
face terminologies: facilitating direct entry of clinical data
into electronic health record systems. J Am Med Inform Assoc
Future consideration of these maps may involve integra- 2006, 13(3):277-288.
tion of the SNOMED CT terminology into the application 3. Chute CG, Elkin PL, Sherertz DD, Tuttle MS: Desiderata for a clin-
ical terminology server. Proc AMIA Symp 1999:42-46.
interface or in a cross-referencing table. It may be that 4. Elkin PL, Brown SH, Carter J, Bauer BA, Wahner-Roedler D, Berg-
exact concept matches will have the most immediate strom L, et al.: Guideline and quality indicators for develop-
potential for integration. Further investigation, i.e. com- ment, purchase and use of controlled health vocabularies. Int
J Med Inform 2002, 68(1–3):175-186.
paring how many exact concept matches correspond with 5. Spackman KA, Campbell KE, Cote RA: SNOMED RT: a reference
the frequency of clinically used terms in the actual legacy terminology for health care. Proc AMIA Annu Fall Symp
1997:640-4.

Page 5 of 6
(page number not for citation purposes)
BMC Medical Informatics and Decision Making 2008, 8(Suppl 1):S3 http://www.biomedcentral.com/1472-6947/8/S1/S3

6. Office of the Secretary, HHS: HIPAA administrative simplification: stand-

ards or electronic health care claims attachments. Proposed rule. Fed. Regist
70(184):55989-6025. 2005 Sep 23
7. Miller RA, Pople HE Jr, Myers JD: Internist-1, an experimental
computer-based diagnostic consultant for general internal
medicine. N Engl J Med 1982, 307(8):468-476.
8. Spackman KA, (ed), et al.: SNOMED Clinical terms January 2005 release
Northfield IL: College of American Pathologists; 2005.
9. Health Language Terminology Server [http://www.healthlan
guage.com/]
10. Clinical Information Consultancy [http://www1.clininfo.co.uk/
home]
11. SNOMED International Concept model working group
[http://www.ihtsdo.org/about-ihtsdo/collaborative-space/]
12. College of American Pathologists: Attributes used in SNOMED CT.
SNOMED Clinical terms® User's Guide-January 2005 release. Northfield,
IL 2005:38-53.
13. College of American Pathologists: Supporting post-coordination.
SNOMED Clinical terms® Technical Implementation Guide-January 2005
release. Northfield, IL 2005:62-63.
14. International Health Terminoloogy Standards Development
Organisation [http://www.ihtsdo.org/our-standards/]
15. Fung KW, Hole WT, Nelson SJ, Srinivasan S, Powell T, Roth L: Inte-
grating SNOMED CT into the UMLS: an exploration of dif-
ferent views of synonymy and quality of editing. J Am Med
Inform Assoc 2005, 12(4):486-494.
16. Kohler J, Munn K, Ruegg A, Skusa A, Smith B: Quality control for
terms and definitions in ontologies and taxonomies. BMC Bio-
informatics 7:212. 2006 Apr 19
17. Bodenreider O, Burgun A: Aligning knowledge sources in the
UMLS: methods, quantitative results, and applications. Stud
Health Technol Inform. 2004, 107(Pt 1):327-331.
18. Cornet R, Abu-Hanna A: Usability of expressive description log-
ics – a case study in UMLS. Proc AMIA Symp 2002:180-4.
19. Smith B: From concepts to clinical reality: an essay on the
benchmarking of biomedical terminologies. J Biomed Inform
2006, 39(3):288-98.
20. Cimino JJ: In defense of the Desiderata. J Biomed Inform 2006,
39(3):299-306.
21. Elkin PL, Brown SH, Husser CS, Bauer BA, Wahner-Roedler D,
Rosenbloom ST, et al.: Evaluation of the content coverage of
SNOMED CT: ability of SNOMED clinical terms to repre-
sent clinical problem lists. Mayo Clin Proc 2006, 81(6):741-748.

Publish with Bio Med Central and every

scientist can read your work free of charge
"BioMed Central will be the most significant development for
disseminating the results of biomedical researc h in our lifetime."
Sir Paul Nurse, Cancer Research UK

Your research papers will be:

available free of charge to the entire biomedical community
peer reviewed and published immediately upon acceptance
cited in PubMed and archived on PubMed Central
yours — you keep the copyright

Submit your manuscript here: BioMedcentral

http://www.biomedcentral.com/info/publishing_adv.asp

Page 6 of 6
(page number not for citation purposes)