Exploiting the user interface for tomorrow's clinical data analysis

Clinical information systems are primary sources of data for health services and effectiveness research. Access to that data may be hampered by complex database structures, and the quality of research lessened by lack of contextual knowledge about the data. In addition, the integration of data captured in multiple information systems remains a challenge because of differences between the information systems in content and data representation. Traditional data integration uses an Extraction-Transform-Load (ETL) workflow which requires that integration of data be performed in a single defined manner. This poses problems for clinical research because the data integration decisions are typically encoded in the highly technical ETL procedures whereas different clinical analyses may require different decisions on semantic integration, which cannot be accommodated in the ETL process. We propose that the contextual information important for data retrieval and analysis resides in the information system user interface and that this interface can be exploited to improve access to single clinical databases as well as the semantic integration of data across multiple databases. The computational concepts and prototype tools presented here include automated tools for creation of databases based on user interfaces, methods for analysts to directly write queries against an interface resembling the user interface and containing contextual information, and tools that allow analysts to classify data elements from multiple sources in a dynamic manner resulting in semantic integration of those data sources.