Natural language processing (NLP) research has matured to the point of being used successfully in applications ranging from search to information extraction to text analytics such as sentiment. Biomedical NLP has shown great diversity of domains, genres, methods, tasks, and applications e.g.. In spite of burgeoning research, few, if any, published papers describe usability or utility of biomedical NLP applications. NLP has not reached its full potential, and one of the reasons may be that there is a disconnect between the linguistically-motivated annotations and the needs of the end users. A central goal of this proposal is to address that gap. We propose to directly link the clinical research user with automated NLP annotations through a general, iterative process correlating the many layers of annotations with user views of the data. The NLP annotators (auto-annotators) and the user interface will mutually inform each other in order to develop an interactive search application for clinical researchers that increases efficiency in reviewing retrospective patient data. We will perform formative and summative evaluations of our NLP-enabled search application, using four research studies as use cases, and will evaluate the final application on four unseen use cases. The final application will comprise a back end for storing annotations and the associated text, auto-annotators, and user interfaces. The application will be released as a stand-alone application and with plug-ins for existing clinical data repositories so that a researcher with an approved research study can use the tool to search and review the relevant patient records.

This project is funded by the National Library of Medicine.