Informative Provenance for Repurposed Data: A Case Study using Clinical Research Data

Abstract

The task repurposing of heterogeneous, distributed data for originally unintended research objectives is a non-trivial problem because the mappings required may not be precise. A particular case is clinical data collected for patient care being used for medical research. The fact that research repositories will record data differently means that assumptions must be made as how to transform of this data. Records of provenance that document how this process has taken place will enable users of the data warehouse to utilise the data appropriately and ensure that future data added from another source is transformed using comparable assumptions. For a provenance-based approach to be reusable and supportable with software tools, the provenance records must use a well-defined model of the transformation process. In this paper, we propose such a model, including a classification of the individual ‘sub-functions’ that make up the overall transformation. This model enables meaningful provenance data to be generated automatically. A case study is used to illustrate this approach and an initial classification of transformations that alter the information is created.