Fig. 3

Process model. Most entries of the terminology originate from a large amount of de-identified documents that are automatically aggregated into more compact files which are the basis for automatically created concept proposals. Development on de-identified documents that were not aggregated allows to further refine the terminology and to detect quality issues. If required, concepts are mapped to standardized external resources. If subsequent evaluation reveals open issues, refinement of segmentation components or other computational aspects can be requested and a new development iteration starts. When all components perform sufficiently, the final information extraction component is deployed and populates a clinical data warehouse