gms | German Medical Science

Introduction: Data exchange between clinical institutions is currently very limited, as most hospital information systems (HIS) possess proprietary data models and therefore transfer or comparison of data between different institutions with different data models is not possible. Ideally, all clinical fact data and all elements in the data’s reference models are annotated with codes from international standardized terminologies, thus facilitating the comparison of data elements amongst HIS at different institutions. When running queries for patient samples, a query should be able to be parametrized using the same terminology IDs, which are defined in the value sets of their reference models. However, in all major existing clinical query languages known to the authors (AQL [1], CQL [2], i2b2 [3], GELLO [4] and Arden Syntax [5]) attributes used in queries always have to be identified by reference model IDs. Queries are thus directly bound to the models. But even if reference model elements were specified by terminology IDs rather than reference model IDs, there are often multiple options to model data objects in local reference models.

Methods: Clinical data objects can be regarded as semantic networks in which the nodes are tagged with terminology codes. A new ideal query language (IQL) should allow querying all possible graphs, each time returning semantically equal results. As the topological structure of models can differ regarding the count of nodes and type of interrelations, the query has to be undetermined about any ontological relationships. Given the reference model of a concrete query system, a graph-matching algorithm has to find a minimum spanning tree of elements that includes all given terminology IDs. When a tree has been identified, a local query language query has to be formulated. The new query has to encompass the reference model elements contained in the spanning tree. Constraints can be added to the new query’s attributes given they are also part of the original IQL query.

Discussion: A problem when searching for the correct minimum spanning tree in a given reference model are potential ambiguities that arise because the set of given terminology IDs matches multiple, semantically different subgraphs:

Different subgraphs bearing the same semantics are a problem of the local HIS rather than the proposed method. Different structures representing semantically equal content should be merged in the HIS into harmonized data structures [6].

A second type of ambiguity is caused by an underspecified set of terminology IDs in the IQL query. It is legitimate that the user resolves this ambiguity manually by adding further discriminating terminology IDs.

The third type of ambiguity are situations, in which multiple subgraphs can only be semantically distinguished by their topology, which cannot be automatically resolved with the proposed mechanisms.

The presented IQL properties as well as the workflow how the IQL interplays with a local query engine may stimulate projects aiming to define a concrete IQL.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.