27-45 JHI_086333.indd

an often existing metadata fi eld 27-45 JHI_086333.indd 34 12/10/2007 2:35:55 PM Process Black 35 Schuurman and Leszczynski Non-equivalent semantic attributes on the premise that data collection practices and an understanding of standardized variable terms are unique to organizations, even in the presence of standardized variable terms. Extended metadata are attached only to fi elds identifi ed – via a fl agging system in the database code or as separate fi les – as requiring additional, extended metadata. Furthermore only pertinent ontology-based metadata fi elds need be fi lled out for fl agged attributes. An example of populated ontology-based metadata is provided in Table 2. The development of ontology-based metadata is profoundly different from the current trend to incorporate ontological context at the model level in GIScience [20, 35, 37, 51]. It is pragmatic, however, in that it presents a vehicle for incorporating use context with data in a manner that is accessible; it requires little re-engineering; and it is intuitively understood by GIS users. These eight fi elds are the basis for extracting the necessary information to export individual fi elds as XML to be used as a boundary object for the purpose of semantic comparison. Database managers have the option of either documenting these extended fi elds in a separate fi le in the tradition of conventional metadata, or at the level of the database (Figure 1). The second alternative allows extended metadata to be stored directly with the data themselves in separate tables linked to fl agged attributes via relational keys. The advantage is that the descriptive information travels with the data. Both these approaches conform to existing data models: either a separate fi le sits on top of the database, and does not interrupt it; or the database fl agging system integrates seamlessly with relational data structures. In both instances, the extended metadata are encoded as XML: either the metadata are exported as such from the database, or the metadata fi le is converted/stored in XML format (most spatial metadata editors have an XML option). Encoding semantic context We construct OWL ontologies as formal representations of semantic concepts in perinatal databases using Protégé (Stanford Medical Informatics) [50]. Each database is ontologically modelled. Semantic comparison is determined via merging ontologies on the basis of a 1:1 mapping between local ontologies. While ontology development is centralized in our approach, it nevertheless affords the high degree of interoperability ‘as mediation between distributed data in [local ontology] environments’ [46, p. 35]. We do not, however, operationalize any of the tools identifi ed by Choi et al. [46] for semi-automating the mapping process. Instead mapping is guided by the user – akin to a supervised classifi cation of remotely sensed imagery – and is executed directly in Protégé. This is particularly salient in the context of local ontology mapping, which is identifi ably more complicated than that between an integrated global ontology and multiple local ontologies because, in the latter, mapping rules are more easily defi ned as all mapping is unidirectional to an intermediate context of standardized defi nitions contained in the global ontology [46]. Moreover, local–local mapping is much more scaleable and thereby more easily streamlined with the web; indeed Choi et al. [46] identify the primary application of local ontology mapping to be the Semantic Web. Because a global ontology represents a semantic ‘least common denominator’, however, the immediate limitation is a loss of semantic granularity. Thus in the process specifi ed herein, the user both authors mapping operations and selects which of the proposed operations, suggested by the system on the basis of auto-detected class similarity, to execute. Merging allows identical concepts, such as ‘baby’ and ‘infant’ in respective databases, to 27-45 JHI_086333.indd 35 12/10/2007 2:35:55 PM Process Black 36 ••••• Health Informatics Journal 14 (1) Ta b le 2 O nt ol og yba se d (e xt en de d) m et ad at a fo r ST IL LB IR TH a nd F ET A L D EA TH fi e ld s in p er in at al d at ab as es f or t w o pr ov in ci al j ur is di ct io ns i n C an ad a: B rit is h C ol um bi a (B C ) an d N ov a Sc ot ia ( N S) . ‘S am pl in g m et ho do lo gi es ’ an d ‘m ea su re m en t sp ec ifi ca tio n’ w er e no t in cl ud ed a s ex te nd ed m et ad at a fi e ld s be ca us e th ey w er e no t ap pl ic ab le in t hi s ca se (i .e . t he re w as n o pe rt in en t in fo rm at io n fo r th es e fi e ld s)