|
|
|
|
|
|
|
Case study:
Unified Medical Language
System (UMLS) |
|
|
|
History |
|
Overview |
|
Benefits |
|
Limitations |
|
|
|
|
|
|
Started in 1986 |
|
National Library of Medicine |
|
“Long-term R&D project” |
|
Complementary to IAIMS |
|
|
|
|
|
|
Bill Hole |
|
L. Kingsland III |
|
Dan Masys |
|
Alexa McCray |
|
Stuart Nelson |
|
Roy Rada |
|
Rick Rodgers |
|
Peri Schuyler |
|
Brigham & Women’s H. |
|
Carnegie-Mellon Univ. |
|
Columbia Univ. |
|
Lexical Technology, Inc. |
|
Massachusetts General H. |
|
UCSF |
|
Univ. of Pittsburgh |
|
Univ. of Utah |
|
[…] |
|
|
|
|
|
Definition of 3 knowledge sources (1986-88) |
|
Metathesaurus |
|
Semantic Network |
|
Information Sources Map |
|
Building, distributing, and testing (1989-91) |
|
Integration vs. ad hoc development |
|
First release in 1990 |
|
Development of applications (1992-94) |
|
|
|
|
|
|
|
Core vocabularies |
|
anatomy (UWDA, Neuronames) |
|
drugs (First DataBank, Micromedex) |
|
medical devices (UMD, SPN) |
|
Several perspectives |
|
clinical terms (SNOMED, CTV3) |
|
information sciences (MeSH, CRISP) |
|
administrative terminologies (ICD-9-CM, CPT-4) |
|
standards (HL7, LOINC) |
|
|
|
|
|
Specialized vocabularies |
|
nursing (NIC, NOC, NANDA, Omaha, PCDS) |
|
dentistry (CDT) |
|
oncology (PDQ) |
|
psychiatry (DSM, APA) |
|
adverse reactions (COSTART, WHO ART) |
|
primary care (ICPC) |
|
Knowledge bases (AI/Rheum, DXplain, QMR) |
|
|
|
|
|
|
|
|
|
|
Two-level structure |
|
Semantic Network |
|
134 Semantic Types (STs) |
|
54 types of relationships
among STs |
|
Metathesaurus |
|
800,000 concepts |
|
~10 M inter-concept
relationships |
|
Link = categorization |
|
|
|
|
|
|
|
|
Broader scope |
|
Extended coverage |
|
Finer granularity |
|
Unique identifier |
|
Synonym terms clustered into concepts |
|
Additional synonyms |
|
Additional hierarchical relationships |
|
Semantic categorization |
|
|
|
|
|
|
|
|
Licensing mechanism |
|
|
|
Too much information |
|
|
|
Not enough information |
|
|
|
|
|
Free UMLS registration |
|
4 levels of restriction |
|
L0 (~55%) must acknowledge NLM, no
redistribution |
|
L1 (~6%) must negotiate for translation |
|
L2 (~.1%) must negotiate for creating health
data |
|
L3 (~39%) must negotiate for any production
use |
|
Possible license fees for certain vocabularies |
|
MetamorphoSys helps subset by source |
|
|
|
|
|
Huge |
|
1.5 M unique English strings |
|
775,000 concepts |
|
Over 10 M interconcept relationships |
|
Complex two-level structure |
|
Metathesaurus |
|
Semantic Network |
|
Steep learning curve |
|
|
|
|
|
Update |
|
Frequency |
|
Mechanism |
|
Lack of coverage |
|
Major sources |
|
Major subdomains |
|
Terminology vs. Ontology |
|
|
|
|
|
|
|
Terminology integration is a step towards
interoperability |
|
Clusters of synonyms from different sources |
|
Paths between terms from different sources |
|
|
|
|
However, interoperability requires more than
loosely aligned terminologies |
|
The UMLS does not claim to be an ontology |
|
The UMLS is, however, a resource for acquiring
biomedical ontologies |
|
|
|
Notes