Creating Reusable Annotated Corpora with the Clinical Document Architecture

Manual annotation of clinical documents creates reference standards to train and evaluate natural language processing (NLP) systems, but can also be used in other aspects of patient care and research. Manual annotation is costly and time-consuming and is usually done for specific use cases. We present a practical approach to the storage of annotations along with text using the Clinical Document Architecture (CDA). We describe the types of annotations commonly made on clinical text and show how they map to CDA elements. This provides standard, reusable documents that allow multiple users to modify and add annotations for new or expanded clinical use cases. As clinical texts become more widely available and NLP systems are applied to extract information from the text, annotated corpora will become an increasingly essential resource. We demonstrate how the CDA provides an appropriate mechanism for storing annotated corpora and describe how it supports interoperability and reusability.