The Significance of SNODENT MIE 2005, Geneva

1
The Significance of SNODENTMIE 2005, Geneva
Louis J. Goldberg, Werner Ceusters, John Eisner,
Barry Smith

• Dr. Werner Ceusters
• European Centre for Ontological Research
• Saarland University
• Saarbrücken - Germany

2
SNODENT

• Structured NOmenclature of DENTistry
• designed as a diagnostic companion to the Current
  Dental Terminology (CDT) treatment codes of the
  American Dental Association (ADA)
• A collection of 6491 terms, partly (4900) taken
  from SNOMED, partly (1600) contributed by ADA
• Integrated in SNOMED-CT starting with the January
  2004 version

3
Quality of SNODENT

• Has the potential to become a de facto standard.
• But is it of good enough quality ?
• To answer that question
• We compared it with SNOMED
• We applied quality checking methods that have
  been used with succes in other checking other
  systems
• SNOMED-CT (MEDINFO 2004)
• NCI-Thesaurus (Meth Inform Med 2005)

4
Materials

• SNODENT as a flat list of
• 6491 unique enomens, each of them associated
  with one out of
• 3863 unique termcodes
• The termcodes are suspiciously similar to SNOMED
  concept codes
• D5-10000 Dental disease, NOS
• D5-10000 Disease of teeth, NOS
• D5-10000 Tooth disorder, NOS
• F-51540 Expectoration of bloody sputum
• F-51540 Expectoration of hemorrhagic sputum
• Thus we were interested to find out how that
  version of SNODENT compared to relevant content
  of SNOMED-CT (July 2003 version)

5
Methods

• Simple analysis and comparisons
• face validity of SNODENT-terms with respect to
  the associated codes (synsets)
• There was no documentation on what the enomen
  termcode relationship in SNODENT represents !
• Overlap of terms/enomens and conceptcodes/termcode
  s in both systems
• Status of the SNODENT-terms if it were
  SNOMED-terms
• SNODENT-graph extraction from SNOMED
• Re-classification of that graph using our
  MEDINFO-2004 algorithm

6
SNODENT-graph extraction from SNOMED

• For each enomen of SNODENT extract all
  SNOMED-concepts to which that term is associated,
  and for each such SNOMED-concept extract all
  concepts on any path to the top, going over both
  isa and associative relationships

ISA
associative
7
The re-classification algorithm

• Automated post-coordination using version space
  theory
• Guided creation of
• Most specific generalisations
• If there is a concept with properties P1 and P2,
  then create a concept that is defined by these
  properties
• Most generic specialisations
• Reclassify all existing concepts having the
  properties P1 and P2
• Process guided by existing subsumption hierarchy
• Inspect suspicious configurations

8
Suspicious configurations
The presence of only one generated concept in a
list of the concepts subsumed by a given concept
The presence of only one existing subsumed
concept next to a list of generated concepts for
the same subsumer
The presence of a pre-existing concept that is
subsumed by a generated concept without any other
additional relationships from the pre-existing
concept to another one.
(xxx)
(xxx)


XXX
XXX
XXX


XXX


XXX

9
Problems in the Calibration of SNODENT and SNOMED
(1)

• 618 (9.52) of SNODENT terms,
• involving 208 (5.38) termcodes,
• point to concepts in SNOMED-CT that have a
  watch out status, distributed as follows
• retired 86,
• duplicate 15,
• ambiguous 517.
• 1203 (18.53) SNODENT terms point to SNOMED-CT
  concepts that are labelled as active, but that do
  not have (in SNOMED) the term used by SNODENT.
• Hence
• if SNOMED-CT would be taken as the gold standard,
  18.53 of the SNODENT terms must be considered to
  be inappropriate.
• If SNODENT would be taken as gold standard,
  SNOMED-CT would lack 18.53 of the accepted terms
  in SNODENT.

10
Problems in the Calibration of SNODENT and SNOMED
(2)

• 368 (5.67) SNODENT terms are not found in
  SNOMED, although the corresponding concept does
  exist in both systems (and with the same code).
• Differences are due to
• use of adjectival form,
• use of a determiner,
• use of NOS,
• eponyms,
• spelling variants.
• SNODENT enforces just a single meaning for terms
  that are in and of themselves polysemous, but
  SNOMED allows terms to be used in a variety of
  meanings.
• 437 (6.73) SNODENT terms are used in SNOMED with
  different meanings, the majority reflecting a
  (systematic) oddity of SNOMED rather than of
  SNODENT.

11
Examples of single terms in SNODENT used with
plural meanings in SNOMED
12
What does the SNODENTenomen termcode relation
captures ?
Reference to the same universal ?
D5-10000 Dental disease, NOS D5-10000 Disease of
teeth, NOS D5-10000 Tooth disorder,
NOS F-51540 Expectoration of bloody
sputum F-51540 Expectoration of hemorrhagic
sputum
13
What does the SNODENTenomen termcode relation
captures ?
Reference to roughly the same universal
? (differences not being relevant for dentists)
F-A3692 Adverse taste perception F-A3692 Chorda
tympani disorder F-A3692 Dysgeusia F-A3692 Neuro
logic unpleasant taste F-A3692 Parageusia F-A369
2 Perversion of sense of taste F-A3692 Primary
taste disorder
T-53120 Dorsal surface of anterior two-thirds of
tongue T-53120 Dorsal surface of
tongue T-53120 Dorsum of anterior tongue
14
Failure to apply soundontological principles
D5-22070 Cheilodynia D5-22070 Painful lips
D5-10578 Sensitive dentin D5-10578 Tooth
sensitivity
It is not because patient with cheilodynia
refers to the same entity as patient with
painful lips, that cheilodynia and painful
lips refer to the same entity !
15
Results of graph extraction and reclassification

• 7500 SNOMED-CT concepts were extracted (2)
• ( Original SNODENT termcodes 3863 )
• 1081 new concepts generated through
  reclassification
• I.e. 14,4 increase
• Re-classification of SNOMED-CT in total showed
  only a 6 increase (MEDINFO 2004)
• ? Dentistry poorly worked out in SNOMED

16
Example of suspiciousconfiguration

• XXX-10015676XXX
• DEVELOPMENTAL ABNORMALITY OF TOOTH SIZE AND FORM
• CONGENITAL ANOMALY OF TEETH (DISORDER) 79723009
• MOON'S MOLAR TEETH (DISORDER) 109436001
• XXX-10007459XXX
• SUPERNUMERARY CUSP (DISORDER) 234954009
• SUPERNUMERARY ROOTS (DISORDER) 12270007
• STRUCTURE OF CROWN OF TOOTH (BODY STRUCTURE)
  75628001
• DISEASE OF HARD TISSUES OF TEETH (DISORDER)
  46557008
• XXX-10007569XXX
• MOLAR TOOTH STRUCTURE (BODY STRUCTURE) 76928009
• XXX-10009937XXX
• TUBERCULUM PARAMOLARE (DISORDER) 78305006
• TAURODONTISM (DISORDER) 51744007
• DENS EVAGINATUS (DISORDER) 63691004
• MICRODONTIA (DISORDER) 32337007
• PEG-SHAPED TEETH (DISORDER) 29553002
• GENERALIZED MACRODONTIA (DISORDER) 93463007
• MACRODONTIA (DISORDER) 71485000

17
Example of underspecification

• SNOMED currently associates tooth finding only
  with the locations
• digestive structure and
• oral cavity structure
• but not with the locations
• tooth structure and
• jaw region structure.

18
Work to be done by CAP/ADA

• assessing which generated concepts should be
  included in an enhanced version of the SNODENT
  codes and which should be excluded
• assigning a fully specified name to the included
  concepts
• subjecting the differentiating criteria which led
  to the excluded concepts to a critical analysis
  and revising them accordingly
• verifying for the included concepts that they
  subsume all concepts that they should subsume.
• Estimated effort
• the 1081 generated concepts themselves subsume
  between 1 and 37 concepts.
• Many of the generated concepts that subsume only
  one concept have a meaning identical to that of
  the existing SNOMED-CT concept.

19
Conclusion

• SNODENT as a flat structure isnt very helpful
• SNODENT in snomed-ised form is better, not
  because the structure is good, but because the
  structure allows you to find mistakes better
• SNOMED currently covers SNODENT not good enough
• The part of SNOMED dealing with dentistry is
  structurally worse organised than the entire
  structure