Automated Mapping of Observation Archetypes to SNOMED CT Concepts - Foundations on Natural and Artificial Computation - page 552

- The Metathesaurus includes many relations of synonymy (more than SNO-
MED CT), which increases the chances of successful searches. For example,
the term Cerebrovascular Stroke is not covered in SNOMED CT. However,
this is a synonym in UMLS.
- Archetypes designers often include words in texts of the nodes that hinder
to find clinical concepts in terminologies. For example: Description of Skin
Color . In these cases, the tool MetaMap [11] is a very useful tool as it is able
to extract the clinical concept, Color of Skin ,ofthetext.
3 Methods
We have developed and tested several mapping algorithms, classically used in on-
tology matching, to bind some terms included in the archetypes to the SNOMED
CT. Our algorithms are focussed on mapping archetype terms to pre-coordinated
SNOMED CT concepts. Several lexical and semantic techniques were applied to
achieve the term mapping.
3.1 Preprocessing
In order to evaluate our approach, a set of test archetypes were selected from
the repository created by NHS [12]. An ADL parser from the openEHR java
reference implementation [13] was used to remove the details of archetype model
syntax and preserve only the data hierarchy. In addition, only data nodes of
the hierarchy with clinical meaning were selected for the mapping process input.
Specifically, only the nodes Element , Cluster , Point Event and Interval Event
were mapped to SNOMED CT concepts. For example, in the archetype in (Fig. 1)
the nodes: at0000 , at0004 , at0005 , at0006 , at0007 , at0008 , at0013 were selected.
The final result of parsing is a tree with different types of nodes: one root
node, which contains the general observation recorded in the archetype ( Res-
piration in Fig. 2), several element nodes which contain patient data related
to the general observation (e.g. Rate and Rhythm in Fig. 2) and several value
nodes which contain the possible values that the elements can take (e. g. Regular ,
Irregular , ...in Fig. 2). The parser extracts the following information for each
node: the identifier ( atxxxx ), the text of section term definitions, the SNOMED
CT concept identifier (if any) of section term binding of the ontology and the
type of node.
Once the trees are obtained by the parser for a set of archetypes, our method
traverses these trees with the aim of mapping tree nodes. Firstly, the root node
is mapped by applying a combination of matching methods, and then the rest
of the nodes are mapped.
3.2 Mapping of the Root Node
In this section, different methods to mapping the root node of archetypes are
detailed.