An archetype is a formal model of a domain concept, expressed as a set of constraints based on a reference model and used in a particular domain. In clinical domains, an archetype is defined as a formal model of a clinical concept. Archetypes are considered by many researchers a fundamental tool for the achievement of semantic interoperability of electronic healthcare records. To date, different libraries of archetypes and a series of tools have been developed for archetype-based standards like ISO 13606 and openEHR. However, despite the importance of the semantics of the clinical content and the demonstrated usefulness of ontologies and semantic web technologies, the available platforms for managing archetypes do not exploit them. In this paper we present our archetype management system, which enables the semantic management of archetypes, performing a series of activities based on their OWL representation. The activities include the transformation of archetypes between standards, their validation or the recommendation of appropriate learning contents.
The semantic interoperability of clinical information remains being
a major goal to achieve in the medical informatics field. Among
the Electronic Healthcare Record (EHR) standards envisaged
for making it possible, the European ISO 13606 standard
(2012), defined by the Technical Committee 251 (CEN/TC 251),
the
Archetypes include bindings to terminologies, which provide
the clinical meaning of the terms of the archetype. Each clinical
institution might develop their own archetypes, thus mechanisms
for managing and assuring their quality are needed. In
Despite the growing importance of archetypes, there are not
many systems for their management. In
In recent years, the bio-ontologies community has developed more than 200 biomedical ontologies, terminologies and controlled vocabularies, most of them being available in the Bioportal website (http://bioportal.bioontology.org/). Such knowledge might play an important role in the semantic activities associated with archetypes and whose aim is the achievement of semantic interoperability. However, to date, such set of ontologies has not been used for such tasks, and its application would also contribute to more effective translational research processes.
In addition to this, the research carried out by our group on the use of ontologies for representing archetypes has produced a series of methods for importing, exporting, validating, annotating and searching archetypes using OWL technologies. In this paper, we will try to provide an integrated solution that combines such methods and the Bioportal for providing semantic archetype management methods, which are available in our archetype management system (ArchMS). ArchMS adopts a semantic approach in which most of the activities are performed over the OWL representation of archetypes and it is able to work with both ISO 13606 and openEHR standards. In order to perform the semantic tasks, the system will combine its internal semantic representation of archetypes with the ontologies available in the Bioportal by using the NCBO web services. The availability of such flexible ontological infrastructure makes possible to represent and exploit all the aspects related to archetype in a common technological, semantic environment. This will make easier the adaptation of the platform to the needs of particular groups of users, since each group might annotate and classify the archetypes according to their own clinical needs. The semantic methods would then exploit such semantic dimensions of interest for providing them with the appropriate archetypes for performing such taks. Given the referred common technological space, those methods would be able to combine governance and terminological annotations of the archetypes to get relevant and consistent results.
ArchMS also addresses the recommendation of personalized
learning contents based on archetypes. According to
ArchMS (http://sele.inf.um.es/archms) can be used
by clinical institutions to develop its own archetypes repository
or as a central common one. Its purpose is to facilitate the local
management of archetypes and to import new ones from other
repositories. The architecture of the system is depicted in Figure 1.
It provides functions for finding similar archetypes, validating them,
and defining search conditions. For this purpose, the system will
group different services and tools which will support the execution
of those semantic activities.
We described now such activities:
• Archetype Representation: ArchMS exploits OWL-based
archetype representations as both individual and classes. On
the one hand, the individual-based representation is the
result of the development of a series of OWL ontologies for
representing the structural semantics of archetype standards.
As a result of this interpretation process, ontologies for
the reference and archetype models of both ISO 13606
and openEHR were developed (see
The persistence layer includes four different repositories: • The knowledge base stores the OWL archetypes which will allow the system to use the semantic representation for different purposes. This repository was developed using Jena (http://jena.sourceforge.net/). • The Archetype Lucene Index refers to the index directory created by the Apache Lucene API, a software library which provides full text indexing and searching capabilities. This repository is mainly used by the search modules. • The relational database contains the archetype information which is continuously accessed by the system, such as its archetype identifier, language, lifecycle state, etc. • The learning contents repository is a semantic repository of SCORM learning contents. This knowledge repository was developed using also the Jena Semantic Web Framework. 3
In this section we will provide a more detailed explanation of how ontologies are used in ArchMS. They are mainly used with the following purposes: • Modelling framework for the domain knowledge; • Background knowledge for supporting classification processes; • Semantic context for calculating similarity. The ontologies used in ArchMS are written in OWL and are either ontologies developed by our group or reused from the Bioportal. 3.1
As it has been described in Section 2, ontologies for representing archetypes and EHR standards (EHR ontologies) have been produced. Such ontologies are the semantic backbone of most ArchMS functionality. Despite ADL being the input format for archetypes in our system we will use their ontological representation in order to perform the semantic activities implemented in ArchMS. More specifically, we will use two different ontological representations, individual-based and class-based representations, depending on the tasks that will be performed. In this way, when the archetype is imported into the system it will be transformed into both representations.
On the one hand, the individual-based representation of the archetype will be used in order to recommend annotations for it by using similarity based searching methods, for generating web applications and for making semantic queries. On the other hand, the class-based representation will be used for validating the correctness of the archetype during the import process.
Apart from managing archetypes, ArchMS also manages learning
contents, in the form of learning objects. According to
The semantic annotations are defined over the set of ontologies managed by ArchMS. Therefore, both archetypes and learning contents are classified using the same knowledge areas, which permits developing effective recommendation methods. Such supporting ontologies may have different origins in ArchMS. First, users can upload their own OWL annotation resources by using the corresponding option in ArchMS. Second, when an archetype is imported, its term bindings are processed. Term bindings are annotations of the archetype content using clinical terminologies and ontologies. In case such terminologies or ontologies are not found in ArchMS, they are searched in the Bioportal by using the NCBO Web Services. If such resource is available in the Bioportal in OWL format, then it is imported into ArchMS.
We have mentioned that ArchMS suggests annotation terms based on the similarity between archetypes. However, it also suggests annotation terms from Bioportal resources. For this purpose, NCBO Web Services are used with text pieces of the archetype. By proceeding in this way and leading towards Bioportal resources-driven annotation processes, we pursue to generate sets of interoperable archetypes annotations. 3.3
ArchMS also uses ontologies for calculating two different types
of semantic similarity. On the one hand, the similarity between
archetypes is calculated by finding how similar two individuals
of the same ontology are. The semantic similarity functions are
based on the ones described in state of the art literature (see, for
instance,
On the other hand, the sets of annotations of both archetypes and learning resources are also compared in order to identify the relevant resources for a given archetype. For each archetype and learning resource, their semantic profiles are obtained by processing its annotations. In this case, the similarity score is based on the taxonomic similarity of the pairs of annotations and only the annotation resource ontologies are used. 4
In this section, part of the functionality provided by ArchMS will be shown by using the openEHR archetype for non-TNM staging scores for the colorectal cancer. 4.1
Once the ADL archetype is submitted to ArchMS its correctness is validated with ARCheck. In case this process is successful, then it is transformed into the individual-based OWL representation and inserted into the repositories. Then, an initial annotation process looks for the most similar archetypes. This would allow reusing annotations from those archetypes. Figure 2 shows the result of such tasks for our running example. In addition to this, the terminological bindings of the archetype are processed to generate new annotations. In this case, since the archetype has an annotation from SNOMEDCT, this annotation is also stored. Besides, the archetype is also processed with the Lucene API to generate information useful for the search facilities. ArchMS offers query facilities for searching existing archetypes. For example, using the semantic search interface the user could query for archetypes stored in ArchMS which contain the keyword cancer. Such query is built by adding constraints to the different properties defined for the concept ARCHETYPE in the EHR ontologies. The definition of this query would be retrieve all the archetypes that: • has description is an ARCHETYPE DESCRIPTION • The details of such ARCHETYPE DESCRIPTION is an
ARCHETYPE DESCRIPTION ITEM • Such ARCHETYPE DESCRIPTION ITEM has a keywords whose value is cancer As a result the tool returns the archetypes ARCHETYPE openEHREHR-CLUSTER.tnm staging 7th.v1, ARCHETYPE openEHR-EHRCLUSTER.tnm staging 7th-breast.v1 and ARCHETYPE openEHREHR-CLUSTER.tumour colorectal staging non tnm.v1. 4.3
Figure 3 shows the annotating interface, which helps the user in the annotation process by making suggestions. ArchMS makes suggestions from both its annotations resources and the ones available in the Bioportal by processing the textual information contained in the description and ontology sections of the archetype. In case such resources do not exist in ArchMS, they can be added by selecting the install option, which is not shown in the figure. If we select SNOMED-CT, the left panel would show the suggestions from such resource. Figure 3 displays suggestions for the keyword metastases. The lower part of the figure shows the existing annotations for the archetype. For instance, the third item, namely, Multiple Malignancy, from SNOMED-CT, was added to the archetype during the import process. This guideline was recommended because the similarity between the archetype annotation Multiple malignancy and the guideline annotations Adenosquamous carcinoma and Carcinoid tumor is greater than the threshold, which for this example was 0.7. Figure 4 illustrates the taxonomic proximity of such concepts in SNOMEDCT.
Archetypes are considered an important element in the achievement of semantic interoperability among EHR systems. So, the design of methods to manage them is fundamental. In this work, we have proposed a system to manage archetypes which take advantages of their OWL representation in order to exploit their semantic relations. The system allows to classify the archetypes, to find similarities between different ones and to check their consistency. The current version of ArchMS integrates different tools developed by our research group in recent years. Some of them were developed for one of the EHR standards managed by ArchMS. However, given that we have methods for transforming archetypes between the standards, all the activities can be performed on both ISO 13606 and openEHR archetypes. ArchMS represents archetypes as both OWL individuals and OWL classes, having different methods for generating such representations. In technical terms, archetypes can be approached as a data object or as a knowledge object depending on the activity to be performed on it. For example, transforming an archetype between standards and checking the correctness of an archetype require a different manipulation and exploitation of the archetypes. In fact, our experience also reveals that given the different semantics of the ADL and OWL (e.g., specialization), the execution of different activities with OWL technologies might require slight changes in the OWL representation. This issue should not be compared with the dual class-individual representation provided by OWL punning. In our case, we aim at keeping such representations at the technical, tooling level, so users could benefit from functionality powered by semantics.
The annotation feature is a useful instrument to support the classification and identification of similarities in archetypes. For this purpose, ArchMS is able to use the Bioportal ontologies, which is in line with our objective of facilitating interoperability. Bioportal ontologies seem appropriate for supporting annotation processes in our system, because our current algorithm makes use of the taxonomic distance between the concepts. However, most of them have limitations in terms of the number of property axioms of the classes, which is a handicap for our property-based similarity factor. The future use and experiences with ArchMS might provide some hints about their appropriateness and their limitations for wide use with archetypes. We have described how archetypes can be used as an instrument for retrieving relevant learning contents. We plan to extend the approach to the data level by using archetyped EHR extracts. This would permit to obtain a more specific set of educational materials for a given citizen. Moreover, it would be interesting to classify resources according to the target audience, that is, professionals or citizens.
This project has been possible thanks to the funding of the Spanish Ministry of Science and Innovation through grant TIN2010-21388C02-02. M.C. Legaz-Garc´ıa is supported by the Fundacio´ n Se´neca through grant 15555/FPI/2010.