caBIO at BOSC 2002 NCICB

1

Use of Mapping for caBIG Standard Vocabularies
Issues to Consider
Margaret Haber Office of Communications National
Cancer Institute
2
Overview

• Mapping Why
• Mapping How
• NCI Metathesaurus example methods
• caBIG Mapping Issues and Questions

3
One Answer is MappingRelating Terminologies for
Effective Data Exchange

• A holistic view of information exchange requires
  broader interoperability, butwhere do we place
  the fences?
• Clinical data, regulatory submissions, discovery
  research?
• Industry agreements, nationally accredited,
  global standardization?

4
The Pillars of InteroperabilityNecessary but not
sufficient

• Common information models across all domains of
  interest
• A foundation of rigorously defined data types
  (metadata)
• A methodology for interfacing with controlled
  vocabularies

5
Interoperability Keys for Terminology

• Use of Industry Standards, where feasible
• Must allow for extensions to core standards
• Specialty terminology remains common
• Mapping is therefore essential
• Conformance with Data Models
• For process (logical models)
• For data flow (messages)
• For data at rest (database design)

6
Mapping has been named as an essential part of
enabling effective interchange between core,
standard terminologies. Complete mapping includes

• An initial high level map performed using
  algorithmic insertion formulas, such as lexical
  matching, followed by human review for accuracy
• Rule-based mapping of non-overlap areas using
  consistent, explicit coding principles
• Actual testing and implementation in systems that
  demonstrate accuracy and effectiveness for data
  capture, transfer, collection and interpretation

7
High Level Map

• Identifies content/concept overlap and gaps
• Reveals issues of semantic equivalence or
  synonymy, differences in definition of terms
• Allows comparative analysis of classification
  systems and structural models
• Provides comparison of content coverage in
  required domains
• Allows system query to quantify comparisons for
  further analysis

8
COSTART vs MedDRA vs SNOMED
9
Rule Based Map

• Often application and system dependent
• Requires domain knowledge and consensus to judge
  appropriate methodology for concept-walking
  between terminologies
• Must grapple with issues of pre- vs.
  post-coordination (full specification vs.
  composition) of terms and concepts
• Terminology assessment and gap analysis
  identifies new concept needs, driving both
  structural requirements for the terminologies and
  the mapping rules between them

10
As an example of some issues that can be involved
. . . .
11
SNOMED CT

• Merger of the College of American Pathologists
  Systematized Nomenclature of Medicine (SNOMED)
  Reference Terminology (RT) and UK Clinical Terms
  Version 3 (CTV3 or Read Codes)
• Broad based clinical terms (CT) with 350,000
  concepts, gt1.37 Million semantic relationships
• Terms of License Five year free distribution in
  US through NLMs UMLS for English and Spanish
  versions including concepts, descriptions,
  relationships, and history. If terminated, the
  last version of SNOMED remains distributable with
  no further updates provided

12
MedDRA (v7.1)

• Description
• International dictionary of medical terms
  including some 65,000 diagnoses, signs
  symptoms, adverse drug reactions, therapeutic
  indications, names/results of laboratory,
  radiological, and other investigations,
  surgical/medical procedures, social circumstances
• Developer
• Under the auspices of the International
  Conference on Harmonization of Technical
  Requirements for Registration of Pharmaceuticals
  for Human Use (ICH) as agreed-upon terminology
  for regulatory reporting
• Terms of Licensing
• Maintenance and Support Services Organization
  (MSSO) supplies MedDRA in various formats,
  multiple levels of subscription

13
MedDRA vs SNOMED Content

• MedDRA content coverage, granularity (level of
  detail) and fully specified (pre-coordinated)
  terms suited to the capture of data relevant to
  adverse event reporting for drugs
• SNOMED has broader/more granular content
  coverage for the representation of information in
  many domains of patient records, potential for
  enabling post-coordination (composition) of
  concepts

14
MedDRA vs SNOMED Structure

• MedDRA hierarchy not is_a but rather groups of
  related terms in a structure designed to maximize
  capture and retrieval of relevant related terms
  these terms may fall into more than one higher
  level grouping
• SNOMED is concept based, with terms and true
  synonyms classified in an is_a
  poly-hierarchical structure it is more reliant
  on composition to fully express granular concepts

15
Example of MedDRA Terminology Links for Toxic
Epidermal Necrolysis
Skin and Subcutaneous Tissue Disorders
Infections and Infestations
Injury, Poisoning and Procedural Complications
Immune System Disorders
SOC
Epidermal and Dermal Conditions
Ancillary Infectious Topics
Chemical Injury, Overdose, and Poisoning
Allergic Conditions
HLGT
Bullous Conditions
Inflammatory Disorders Following Infection
Poisoning and Toxicity
Allergies to Foods, Food Additives, Drugs and
Other Chemicals
HLT
Toxic Epidermal Necrolysis PT
PT
Drug Eruption Lyell Syndrome Type
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
LLT
Drug Eruption Lyell Syndrome Type
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
Necrolysis Epidermal Toxic (Lyell Type)
16
Example of SNOMED Terminology Links for Toxic
Epidermal Necrolysis
Chemical compound
Barbitone chemical
Chemical product
Chemical categorized structurally
Chemical suspension
Antipyrine
Has causative
Pyrazole derivative

• Toxic epidermal necrolysis Synonyms Lells
  toxic epidermal necrolysis, supdpidermal type
  TEN Toxic epidermal necrolysis Lyells
  syndrome

Dermis
Has finding site
Is a
Is a
Chemical-induced dermatotogical disorder
Non-infectious, vesicular and/or bullous disease
Disease of skin and/or subcutaneous tissue
17
Issues to consider when looking at potential
standard terminologies before undertaking maps

• Terminologies in use for health communications
  reporting must have clear policies and procedures
  for both national and international use
  (scope/extent of license, fees, maintenance)
• Reliable mechanisms for user feedback must be in
  place to ensure requirements for updates can be
  incorporated into the terminology and published
  on a regular basis
• The above requirement (2) becomes more
  challenging with terminologies that have very
  broad coverage or a wide user base

18
Issues to Consider . . .

• Concept capture is only one part of the story
  structure such as hierarchy is critical for
  aggregation and retrieval, and very important
  particularly for reporting applications
• Hierarchies are also important in mapping, to
  indicate the intended meaning of a concept if
  ambiguous, and to provide appropriate terms for
  mapping (up-coding) from more granular
  terms/terminologies. Cross-maps thus require
  comparative structural analysis
• Size of the terminology and depth of hierarchies
  can be a significant issue for applications

19
NCI EVS Goal Integration by Meaning

• Clinical, translational, and basic research
  terminology have overlapping but specialized
  needs, therefore EVS assists to
• Integrate different conceptual frameworks
• Create terminological and taxonomic conventions
  across systems
• Provide terminology for
• Tagging store/transfer/archive data for future
  analysis
• Reasoning limited inferencing about data

20
NCI Metathesaurus

• Filtered UMLS Metathesaurus extended with
  additional required vocabularies
• 930,000 concepts, 2,200,000 terms and phrases
  with definitions
• Mappings among over 50 vocabularies
• Extensive synonymy Over 40,000 terms for
  neoplasms mapped to 7,000 concepts
• Used as online dictionary and thesaurus, for
  mapping and document indexing

21
NCI Metathesaurus (2)

• Minor releases monthly, Major releases twice a
  year
• Provides a mapped overlap and partial
  inter-relation of current versions of NCI
  required vocabularies, ex. The ICDs, MedDRA,
  SNOMED, MeSH (NLM Medical Subject Headings),
  HCPCS (procedures), LOINC (lab values), drug
  terminologies (VA NDF, AOD, RxNORM, Multum, NCI
  Thesaurus drugs, etc.)

22
NCI Metathesaurus Data Profile

• Multiple sources per concept or unit of
  semantic meaning
• Hierarchically organized to reference NCI
  Thesaurus structures when possible alternate
  source hierarchies may also be viewed
• 109 Term Types within concepts such as
  preferred term (PT), synonyms (SY), abbreviations
  (AB), obsolete (OB), brand name (BR) drugs, etc.

23
NCI Metathesaurus Profile (2)

• 135 Semantic Types divide groups of concepts
  into general domains of meaning such as
  Neoplasms, Genes, Pharmacologic Substance,
  etc.
• gt5,600,000 asserted relationships between
  concepts ex.
• Carcinoma Clinically_associated_with Lytic
  Bone Lesion,
• TP53 Gene_associated_with_Disease Breast
  Carcinoma

24
(No Transcript)
25
 
 
 
 
 
 

• For Mapping caBIG Standard Vocabularies
• Is it advisable to offer a caBIG standard
  terminology set based on concept mappings to an
  existing standard terminology?

26
Maybe

• The answer is a qualified yes but its not
  easy, and risks are present.
• Maps are difficult to construct.
• Even carefully constructed maps can convey some
  degree of false synonymy
• Like terminologies, maps need maintenance.

27
Imprecise or False Synonymy

• It may not be possible to find an exact match
  between concepts in two terminologies.
• Often the best one can do is to come close.
• How is the degree of closeness indicated to a
  user?
• For certain applications, e.g., clinical,
  close may not be acceptable.

28
Like Terminologies, Maps Require Maintenance

• A map exists between a pair of terminologies,
  both of which are at specified revision levels or
  versions
• A change to either terminology requires that the
  map be re-verified and possibly redone

29
This Raises Questions

• Who will consistently verify an update maps?
• How committed is the mapping organization to
  this activity?
• How much time is needed to update the map after
  either vocabulary changes?
• What is the status of a mapped terminology
  extension in the interim?

30
One Size Doesnt Fit All

• Theres no single definitive map between a pair
  of terminologies
• Maps must be based on a well-defined use cases
• Different use cases may require different maps
• Example NLM makes multiple maps between CPT
  and ICD to meet varying use cases from providers
  and payers.

31
Vocabularies for Reporting are Especially
Demanding

• Its extremely challenging to substitute an
  alternative terminology for one designed for
  specific purposes.
• FDA requires MedDRA for reporting adverse events
• It would be difficult to code adverse events
  using another terminology and map precisely to
  MedDRA for reporting.
• There is significant potential to undermine the
  integrity of the reports submitted.

32
The Bottom Line

• Mapping is expensive, inherently limited, highly
  particularized, and represents an ongoing burden
• The payoff for a mapping effort must be justified
  by clear operational benefits
• If a mapped terminology becomes a caBIG standard,
  its ongoing maintenance and integrity becomes a
  significant concern for all users of that
  terminology
• The analysis that supports the decision to create
  such a map must consider all the costs and risks
  of doing so.

33
EVS Team Acknowledgments

• NCI Office of Communications
• Margaret Haber
• Larry Wright
• NCI Center for Bioinformatics
• Frank Hartel
• Sherri de Coronado
• Gilberto Fragoso
• Contractors
• Apelon, Inc. Northrop Grumman, Inc.
• Aspen, Inc. Kevric / IMC
• MSD J. Oberthaler Consulting
• SAIC Protégé/SMI
• Collaborations External Reviews
• NCI caBIG, CTEP, DTP, DCP, DCCPS, MMHCC, etc.
• NIH NLM, NHLBI
• Govt FDA, VHA, CDC, DoD, NASA
• Other CAP/SNOMED, AFIP, HL7, CDISC, MGED, W3C