=Paper= {{Paper |id=None |storemode=property |title=OWL-based Reasoning Methods for Validating Archetyped Clinical Knowledge |pdfUrl=https://ceur-ws.org/Vol-674/Paper150.pdf |volume=Vol-674 |dblpUrl=https://dblp.org/rec/conf/ekaw/TortosaF10 }} ==OWL-based Reasoning Methods for Validating Archetyped Clinical Knowledge== https://ceur-ws.org/Vol-674/Paper150.pdf

OWL-based Reasoning Methods for Validating Archetyped
Clinical Knowledge ∗

Marcos Menárguez-Tortosa Jesualdo Tomás Fernández-Breis
Departamento de Informática y Sistemas Departamento de Informática y Sistemas
Facultad de Informática Facultad de Informática
Universidad de Murcia, Spain Universidad de Murcia, Spain
marcos@um.es jfernand@um.es

ABSTRACT archetypes based on semantic web technologies in order to
Most advanced Electronic Healthcare Records architectures perform such semantic activities [5]. However, that represen-
represent clinical knowledge by means of archetypes. Con- tation requires much implementation work for validating the
sequently, guaranteeing the correctness and consistency of knowledge of archetypes and dealing with EHR extracts. In
such archetypes becomes crucial for the success of those ar- [4] a representation that allows for processing EHR extracts
chitectures. In this work, we present a method that uses is proposed, but quality assurance and validation methods
OWL and reasoners for evaluating the consistency of the were not provided.
archetypes. This method has been validated through its ap-
plication to the openEHR archetype repository, which is the In this work, an OWL-based representation of archetypes
largest available one nowadays. The results of this validation that makes it possible to accomplish validation and qual-
are also reported in this study. ity assurance requirements is presented. The methods for
checking the consistency of archetypes will be supported by
reasoners, which will be in charge of helping to identify the
1. INTRODUCTION wrong definitions in the archetypes.
The lifelong clinical information of any person supported
by electronic means configures his Electronic Health Record
(EHR). Most advanced EHR architectures and standards are 2. ARCHETYPES
based on the dual model-based architecture [1] (OpenEHR, Archetypes are detailed and domain-specific definitions of
ISO EN 13606), which defines two conceptual levels: (1) clinical concepts in the form of structured and constrained
reference model; and (2) archetype model. The reference combinations of the entities of the reference model. The ISO
model defines the set of entities that form the generic build- EN 13606 and openEHR communities specify them using the
ing blocks of the EHR. On the other hand, archetypes define Archetype Definition Language (ADL). Next, an excerpt of
how to represent clinical concepts in the form of structured the definition of an openEHR archetype for working with
and constrained combinations of the entities contained in check lists in healthcare is shown.
the reference model, so knowledge in the EHR domain is
defined at this level. -- A health oriented check list
EVALUATION[at0000] matches {
data matches {
-- Tree
Archetype-based EHR domain knowledge plays a fundamen- ITEM_TREE[at0001] matches {
items cardinality matches {0..*; unordered} matches {
tal role for the achievement of semantic interoperability in -- Question group
healthcare [2]. In addition to this, the requirement of formal CLUSTER[at0002] occurrences matches {1..*} matches {
items cardinality matches {0..*; unordered} matches {
methods for validating the design and content of archetypes -- Question
CLUSTER[at0003] occurrences matches {1..*} matches {
has been identified in [3]. So far, very few archetypes au- items cardinality matches {1..2; ordered} matches {
ELEMENT[at0004] occurrences matches {1..*} matches {*}
thoring tools implement techniques for assuring the quality -- A comment on the answer
of archetypes and none includes a knowledge-based repre- ELEMENT[at0005] occurrences matches {0..1} matches {
value matches {
sentation of archetypes in order to perform semantic activ- DV_TEXT matches {*}
}
ities. In previous work, we addressed the representation of }
...
}
∗This work has been possible thanks to the Spanish Ministry
of Science and Innovation through grant TSI2007-66575-
C02-02.
Concepts in archetypes are identified by an id (e.g., at0003)
and characterized by their occurrences, that is, the number
of instances that can be part of the association to which they
belong. Multivalued attributes may be restricted in differ-
ent ways: cardinality, order and uniqueness (unique) of
the instances. Finally, an archetype can be defined as the
specialization of another one.
3. OWL REPRESENTATION In order to provide detailed information about the causes of
In order to check the correctness of the clinical archetypes the inconsistency, support classes, that allow the isolation
based on OWL reasoning, we should be capable of repre- of each archetype constraint, are defined. For instance, a
senting archetypes in OWL. As it has been mentioned, ar- class is generated for defining the constraint of maximum
chetypes are built by defining restrictions over the entities of cardinality for the attribute items of CLUSTER_at0003. The
the EHR reference model, usually expressed in UML. There- definition of these classes for the specialized archetype would
fore, an OWL-based representation of the reference model is be similar, except for the addition of the corresponding an-
needed. Otherwise, the validation of the archetypes would notation to define its parent archetype.
not be possible. This is not a difficult step since this repre-
sentation can be obtained by applying the following simple
Class: MAX_CARDINALITY_COLLECTION_CLUSTER_at0003_items
rules: (1) UML classes are transformed into OWL classes; EquivalentTo: MAX_CARDINALITY
(2) attributes are represented as OWL properties and a Sub- and id value "COLLECTION_CLUSTER_at0003_items"
ClassOf axiom is included in the class in which it has been and max_cardinality_value only xsd:int[ <= 2 ]
defined; (3) inheritance relations are transformed into Sub-
ClassOf axioms and sibling classes are defined disjoint; and 5. VALIDATION AND TOOLING
(4) multivalued attributes are represented by means of a The OWL-based method for checking the consistency of ar-
class which makes use of properties for defining the order, chetypes has been implemented in the tool Archeck that is
repetitions, cardinality and range of the values of the in- available at http://miuras.inf.um.es/archeck. The tool has
stances. been implemented in Java and makes use of the openEHR
Java tools. Ontologies are processed with the OWL API and
An archetype defines a clinical concept by constraining an we have used Pellet and Fact++ for the reasoning.
entity of the reference model. Such concept represents a spe-
cialization of that reference model’s entity. Those restric- Our experimental validation has used the archetypes avail-
tions are applied to the attributes defined for each entity: able in the openEHR repository. The validation of this
range, cardinality, and so on. In this way, each restricted repository has reported that 14 over 931 archetypes are in-
entity is defined by means of an OWL class which contains consistent. In terms of reasoning time, the results obtained
the definition of the corresponding constraints. Most con- with Fact++ are better than with Pellet. The complete re-
straints are applied to multivalued attributes. These con- sults of this experiment are available at the referred website.
straints are defined over the collection class that represents
the range of the property. This is the case of the attribute
items of CLUSTER[at0003] that is shown next.
6. CONCLUSIONS
This work proposes a representation of ADL clinical arche-
types as OWL classes in order to check the correctness of
Class: CLUSTER_at0003 their definitions by using OWL reasoners. The approach
EquivalentTo: ARCHETYPED_CLASS
and id value "at0003" has been implemented in the Archeck tool and has been val-
and CLUSTER idated using the openEHR archetype repository. The tool
and op_items only COLLECTION_CLUSTER_at0003_items has proved to be useful since a number of archetypes have
Class: COLLECTION_CLUSTER_at0003_items been found inconsistent in that repository and the validation
EquivalentTo: COLLECTION errors have been reported. Archeck will also be deployed as
and id value "COLLECTION_CLUSTER_at0003_items" a web service in order to integrate the validation process in
and elements min 1 ITEM
and elements max 2 ITEM other tools.
and elements min 1 ELEMENT_at0004
and elements max 1 ELEMENT_at0005
and ordered value true 7. REFERENCES
and unique value false [1] T. Beale. Archetypes: Constraint-based Domain
Models for Futureproof Information Systems. In
Eleventh OOPSLA Workshop on Behavioral Semantics:
In case of defining an archetype that specializes another one,
Serving the Customer, 2002.
its definition includes an additional annotation in each class.
That annotation indicates the name of the class in the parent [2] European Commission. Semantic interoperability for
archetype that is being specialized, if any. This annotation better health and safer healthcare. deployment and
will then be used in the consistency checking process. research roadmap for europe. ISBN-13 :
978-92-79-11139-6, 2009.
4. CONSISTENCY CHECKING [3] D. Kalra. EHR Archetypes in practice: getting
An archetype is consistent if its set of constraints defined feedback from clinicians and the role of EuroRec. In
over both the reference model and the parent archetype are eHealth Planning and Management Symposium, 2007.
valid. The following types of errors can be identified: incor- [4] O. Kilic, V. Bicer, and A. Dogac. Mapping Archetypes
rect restrictions and incorrect specializations. In this work, to OWL. Technical report, Middle East Technical
such inconsistencies are detected by using a strategy based University, Turkey, 2005.
on OWL reasoners. In the first case, if a class is not consis- [5] C. Martı́nez-Costa, M. Menárguez-Tortosa, J. T.
tent with respect to the reference model, the reasoner will Fernández-Breis, and J. A. Maldonado. A model-driven
find that class unsatisfiable. Specialization errors requires approach for representing clinical archetypes for
the processing of the reasoner results. In this way, a special- Semantic Web environments. Journal of Biomedical
ization is wrong if the reasoner cannot infer a subsumption Informatics, 42(1):150–164, 2009.
relation.