gms | German Medical Science

The authors Blaum and Jarczewski contributed equally to the study.

At the present time, mounting supply of available learning resources in medical education, particularly online, can hardly be used effectively by teachers and learners due to the lack of a simple and systematic mode of access to its contents [1]. Due to the constantly increasing number of available sources, students are confronted with the challenge of identifying, in a sea of possibilities, thematically appropriate, high-quality, and didactically significant material that corresponds to their level of education.

Computer-based learning in particular has experienced a substantial increase of resources. In a recent review in the topic of computer-based virtual patients - a mere subset of the computer-based resources - Cook et al. identified 698 publications on this single issue [2]. At the same time, the PC-based learning is becoming increasingly important with the arrival of the "Generation Y" [3].

Teachers and planners are also being challenged to present (to map) the contextual relationships in the curriculum for which they are responsible [4], a task that until now required only a manual linking of curricular elements.

Notably, development of the Internet towards the Semantic Web (Web 3.0) presents considerable potential for the contextual linking of learning resources, as well as applications in curricular mapping [5], [6].

While at first content of the Internet was essentially determined by a few authors, who provided information to a broad, consuming audience, the current "Web 2.0" is characterized by user-generated content [7]. A common feature in data from both versions is that their information can only be understood and interpreted by humans. Hence, any contextual relation between data requires human interaction.

The term "Web 3.0" (also "Semantic Web") refers to a network data storage system which, through the addition of machine-readable meta-information (per example, via structured vocabularies), allows machines to automatically generate and discover contextual relations between data objects [8]. This innovation requires a machine-readable description of the information contained in data [6], [9].

A suitable type of content description is the use of structured vocabularies. In this case, each individual resource is tagged with the defined terms of a structured vocabulary.

Taxonomies and ontologies provide controlled vocabularies for content description (semantic annotation) of objects, such as, per example, learning resources [10]. Relationship between the terms of a vocabulary are stored within a taxonomy, allowing for an (manually initiated) automated search to identify similar content descriptions among a multitude of learning resources, independent of their location [1], [11], [12].

Relationships between vocabulary terms of a taxonomy are typically monohierarchical, thus each term has exactly one (or no) parent term ("is" - or "is a part of" - relationship), as well as an arbitrary number of subordinate (child) terms. A classic example of a strictly monohierarchical taxonomy is the International Classification of Diseases http://www.dimdi.de/static/de/klassi/index.htm. Conversely, ontologies may also utilize polyhierarchical structures, so that a single vocabulary term may have multiple parent terms. The fundamental difference to taxonomies is that, instead of language constructs, logically defined formalisms are used for the definition of objects and relationships in the real world. Due to this independence from linguistic properties, ontologies represent a robust toolkit for semantic descriptions, and can greatly facilitate the standardization of terminology [13].

A well known example of a polyhierarchical ontology with equivalence relationships are the Medical Subject Headings http://www.nlm.nih.gov/mesh/, which are used to describe publications indexed in Medline. Figure 1 [Fig. 1] clarifies the differences between taxonomy and ontology.

Taxonomies and ontologies are fundamentally suitable to create contextual relationships between objects, even if they are located in separate, independent systems. They may also further assist preparation, if not complete automation, of the necessary – but burdensome - manual linking between different outcome frameworks [14] or learning objective catalogs [15], as they are generally used in a minority of approaches to curriculum mapping.

To the best of our knowledge, generally accepted criteria for description and comparison of structured vocabularies are currently not available.

When choosing an appropriate taxonomy or ontology, medical faculties are faced with a multitude of offers that differ in scope, maintenance, and availability. Aim of this study is to compile existing taxonomies and ontologies used for annotation of medical learning resources and content, to compare those using purpose-built criteria, and to examine their suitability for use in the Semantic Web.

A literature search in the databases of the GMS (German Medical Sciences) http://www.egms.de/dynamic/de/index.htm was initially carried out in June 2011, in order to identify publications describing at least one taxonomy/ontology with current or potential use in the medical field. Searched terms are summarized in Table 1 [Tab. 1], on the left column.

Five structured vocabularies were identified.

The authors extracted criteria used to assess the suitability of a vocabulary for annotation of elements of medical education from these publications in consensus.

These extracted criteria include:

• Vocabulary theme: training formats and methods, as well as biomedical content, should be described.
• Structure: The vocabulary should support content description by a variety of different users, and therefore support synonym and equivalence relationships between terms.
• Language: German-speaking teacher and students should be able to use the respective vocabulary.
• Scope: The vocabulary should be specific enough to differentiate the contents and methods used in medical studies in the German-speaking world; but conversely, not so detailed as to force the users to annotate using irrelevant distinctions.
• Maintenance: In order to remain current, the vocabulary should be updated by a specified institution or organization at least once a year.
• Technology: Data format, availability, and copyright of the vocabulary should allow for a cost-effective use in description of web-based resources and elements of medical curricula.

The second step took place in June and July 2011, and involved an expansion of the systematic literature search in Medline http://www.ncbi.nlm.nih.gov/pubmed and in GMS http://www.egms.de/dynamic/de/index.htm, using 7 supplementary English terms. The additional terms searched are summarized in Table 1 [Tab. 1], on the right column. Based on the abstracts of the search results, all publications describing a formal vocabulary related to medical education were selected for further analysis, and the cited taxonomies and ontologies were identified.

A web search was conducted for each of the structured vocabularies thus identified, and their respective editors were contacted and questioned regarding the actuality, use, size and technical details (such as data format) of their vocabularies.

If the editor of a vocabulary could not be identified, the authors of the publication describing the respective vocabulary were then contacted.

Each structured vocabulary was analyzed in regard to language, hierarchical design, type of cross-referencing, topic/subject area, size, currency, maintenance, format, and availability.

The described search strategy identified 601 publications. Based on the abstracts of these publications, 20 articles were identified describing the development or use of a structured vocabulary in medical education.

Among the 20 articles found, 14 different structured vocabularies were described. Four of these were explicitly designed for description of training formats and methods.

Furthermore, a Metaontology comprising a semantic network of more than 130 different terminologies is available under the "Unified Medical Language System" (UMLS).

The 14 identified vocabularies could be classified into one of three primary applications. These applications include "Training Formats and Methods", "Biomedical Content", and "Administration and Documentation" (see Table 2 [Tab. 2]). In essence, vocabulary structures differ in the method by which their terms are connected: polyhierarchical ontologies include not only the parent-child relationships between terms, but also allow for analysis of equivalency and association references. Conversely, the described monohierarchical taxonomies allow only one type of relationship between its terms, and are at best supplemented by a "synonym" type link. The underlying structure of a vocabulary depends substantially on its purpose or on the content that the vocabulary is designed to describe. Vocabularies primarily developed for documentation and administration tend to follow a monohierarchical structure, while ontologies meant to describe training formats and methods, as well as those describing biomedical content, are usually polyhierarchically structured.

Further essential distinguishing features within these areas of application, such as language and scope, are listed in Table 2 [Tab. 2].

Three of the identified vocabularies designed to describe training formats and methods are mainly in English, such as "Topics for Indexing Medical Education” (TIME), and secondarily in French. The TIME-ontology is designed for description of medical teaching and learning content, while the "Medical Education Taxonomy Research Organization" (METRO) taxonomy is mainly suitable for the annotation of methods and processes of the medical education. The "British Education Thesaurus" (BET) is a general education thesaurus without any specific reference to medicine.

Additionally, an ontology in German language is available through the "Ontology of Bio-Medical Educational Objectives "(OBEO), which is used primarily for the semantic annotation of learning objectives in medical training.

Vocabularies intended for annotation of biomedical content are partially very extensive and/or specialized. Only four of them were designed for a comprehensive medical description, not focusing on a single thematic area, such as anatomy, oncology, or pharmacology (see Table 2 [Tab. 2]), but rather fit to describe extensively a broad area of the biomedical content.

While BET allows extensive annotation of general educational content, the possibility of describing specific medical content with this thesaurus is restricted. As seen in Table 2 [Tab. 2], the other three identified vocabularies (METRO, OBEO, and TIME), which have been designed for description of training formats and methods, are significantly less extensive than vocabularies used for describing biomedical content, or content administration and documentation.

Conversely, the four identified vocabularies for comprehensive description of biomedical content are sometimes very extensive (see Table 2 [Tab. 2]).

Table 3 [Tab. 3] summarizes data on updates and support, data format, and costs for the seven identified ontologies - three for the description of training formats and methods, and four for the description of biomedical content.

Use of structured vocabularies, such as taxonomies and ontologies, enables automatic comparison and exchange of data between institutions and systems [17]. It also allows automatic comparison of multiple sets of information, which automatically identifies overlapping content: a function that would, for instance, enable automatic adjustment of a medical curriculum with any possible outcome frameworks [15] – a particularly interesting feature in terms of accreditation.

Since the vocabularies can in principle be used for the formal content description of any object, they may be used - in addition to traditional resources such as books, media, and programs – to annotate courses, learning objectives, exam questions, or any other form of online resource. As such, they provide a starting point or search results for thematically related objects. A few curriculum mapping implementations utilize this opportunity to show contextual relationships between elements in a curriculum [5], [16]. Users of such modules can be automatically referred to related modules in the same (or another) e-learning platform by machine-readable description of learning module contents in e-learning systems, which is based on taxonomies.

In addition, formalized content description of resources is easy to browse and to maintain [8].

Use of such vocabularies requires consistent annotation of all new objects using a defined vocabulary, which signifies an additional effort when creating new content. Conversely, this eliminates the need for manually linking new content. Semantic description of objects, therefore, can only be effective with an increasing amount of content elements.

No German language based, purpose built taxonomy or ontology for the annotation of resources and medical education content could be identified.

The structured vocabularies available in German - such as OBEO - are not yet specific enough to allow for a meaningful distinction among learning resources, but rather appropriate for the content annotation of patient data (SNOMED-CT) or biomedical publications (MeSH).

Considerable differences in the scope of the identified vocabularies - which resulted primarily from very specific applications, such as the description of teaching formats (METRO) - raise the question of whether their granularity is fine enough for a differentiated annotation of content and methods in medical education.

The ontologies available in English that are purpose built for the domain of medical education, such as TIME and METRO, must first be converted into German. In this case, problems arise in choosing contextually equivalent translations. For instance, the English term "medical education" is equivalent simultaneously to three German terms, namely “medizinischer Ausbildung”, “Fortbildung”, and “Weiterbildung”, which individually refer to the English "graduate", "continuing” and "post-graduate" medical education.

Another problem posed by a translation is the question of whether, and how, a translation can be bound to the original version of the vocabulary in terms of potential developments and updates.

Alternatively, development of an ontology primarily based in the German language is associated with considerable time and effort.

Particularly interesting features among the identified vocabularies, summarized in Table 3 [Tab. 3], include the fact that all are available online, and at least in part, all utilize direct Internet technologies, such as the XML data format. Although they have not been primarily developed for use in the web, they are at any rate still suitable for this purpose. The technical foundations needed towards Web 3.0 are therefore established [9].

As any literature research, this work is limited by the search and selection of the used publications.

Additionally, existence of structured vocabularies not listed in any of the publications of the searched databases cannot be ruled out.

During preparation of this work, the authors were unable to identify any generally accepted criteria for the description and comparison of structured vocabularies. Therefore, the first step was to identify any medical taxonomy described among the publications of the searched databases. This was followed by a publication analysis based on individual criteria - developed by the respective authors of each publication – used to describe the structured vocabularies. By consensus, the authors of this study then extracted the criteria that seemed most adequate for a comparison of taxonomies for medical training. During criteria selection, the authors have sought to generalize the results, making them suitable for use by German-speaking medical faculties. Alternate or additional criteria not considered in this study may be relevant for local use.

Content description (semantic annotation) of medical learning resources provides the ability to identify contextually related items across system boundaries. Thus far, there is no suitable taxonomy or ontology in German that may be used to implement a systematic description of resources for Germany’s medical students, thus limmiting the potential of the Web 3.0 to support curriculum mapping. Possible approaches include the new development of an appropriate structured vocabulary, the translation of existing English-language vocabularies, or a combination of existing vocabularies sections, possibly with a partial translation of English vocabularies.

WB is the head of the Learning Center of the Charité, a member of the Medical Education Society (Gesellschaft für Medizinische Ausbildung - GMA), and of the GMA commitees “Practical Skills” and “Educational Research Methods.” AJ and PS are student tutors of the Learning Center, and students of the Charité. FB is a co-developer of the OBEO. OA is head of the Department of Curricular Management, a member of the GMA, as well as a member of the GMA committee “Educational Research Methods.”

The authors thank Dr. Ullrich Woermann, Bern, as well as two anonymous reviewers for their constructive and collegial critic of the manuscript and Rudi Mörgli for translating the manuscript into English.

The authors declare that they have no competing interests.