gms | German Medical Science

Introduction: ProHTA (Prospective Health Technology Assessment) [1] is a simulation project that aims at understanding the impact of innovative medical processes and technologies at an early stage. To that end, large scale healthcare simulations are employed to estimate the effects of potential innovations [2].

Besides the problems of simulation modeling, validation and optimization in our simulation project, simulation data management is required. This includes semantic integration, since simulation data in our project stems from several heterogeneous sources. Although many techniques exist for automatic integration of data [3], it is still an extensive process.

Howe et al. [4] identified several key barriers to adopting a data management system for science. They argue that the initial effort of designing a data schema is to complicated in most cases. Therefore, we develop a "data first, structure later" approach to reduce the initial effort.

Material and Methods: Since our pool of data sources is likely to expand rapidly, we investigate an integration strategy based on the dataspaces abstraction [5]. This approach allows the coexistence of heterogeneous data sources, which are initially integrated only as far as automatically possible. We investigated how an RDF (Resource Description Framework) [6] triplestore can be utilized to enable the gradual improvement of the degree of integration in a pay-as-you-go manner. Additionally, we analyzed the requirements of our simulation practitioners and developed a prototypical data provisioner for the simulation.

Results: We developed an ontology for storing statistical data in an RDF triplestore [7]. Data and metadata are stored using several generic upper ontologies. We also devised a domain specific query language that assists simulation modelers to load statistical data into simulation models.

This framework has been extended to store data using the automatically imported schema of the data source. The user can query the data utilizing a key word search or by querying the data using information about the source schema. Also, the user is able to view the data with a web front end. This front end also enables the user to add annotations about the underlying semantic concepts of the data. When the annotations are complete, application specific views allow the usage of our domain specific query language to load data into the simulation.

Conclusion: Because of the changing demands of the simulation and the amounts of heterogeneous data, the initial effort of integrating a data source should be minimized. Therefore, our framework enables automatic integration and schema import for sources with known formats. As the data has to be reusable for different simulation studies, the data management system provides the means of adding semantic information to initially integrated data. Additionally, the simulation modeler is able to find and query data easily using a key word search and a domain specific query language. Hence, the effort of using a data provider for simulation input data management is less than the effort of manual data input.

Acknowledgements: This project is supported by the German Federal Ministry of Education and Research (BMBF), project grant No. 01EX1013B.