Narratives and first-person stories allow clinicians to easily document rich information about the patient and their care progress in free-text reports. However, they can get bloated quickly making it hard to browse through them and find relevant information. Using Natural Language Processing (NLP) methods to address this problem is challenging as contextual examples are required to train models for every different use case. Clinicians who can understand medical jargon and create training examples are usually not versed in informatics techniques to be able to use NLP directly. Interactive tools supporting the review and revision of models have the potential to narrow the gap between domain experts and informaticians, making NLP more valuable for clinical applications.
Building on my prior work on applying interactive learning for retrospective research on procedure notes, I am proposing its application for identifying important or relevant portions of clinical notes in electronic medical records. Many current clinical practices involve manually curating information in different summary forms such as sign-out notes, discharge summaries etc. These are time consuming processes and can be improved using interactive NLP methods. By considering an example use case for identifying important sentences in full-text reports for signout note preparation, I would like to study how interactive machine learning methods can be designed to help clinicians build and interactively revise NLP models for their own use.