Annotating Change of State for Clinical Events

Understanding the event structure of sentences and whole documents is an important step in being able to extract meaningful information from the text. Our task is the identification of phenotypes, specifically, pneumonia, from clinical narratives. In this paper, we consider the importance of identifying the change of state for events, in particular, events that measure and compare multiple states across time. Change of state is important to the clinical diagnosis of pneumonia; in the example “there are bibasilar opacities that are unchanged”, the presence of bibasilar opacities alone may suggest pneumonia, but not when they are unchanged, which suggests the need to modify events with change of state information. Our corpus is comprised of chest Xray reports, where we find many descriptions of change of state comparing the volume and density of the lungs and surrounding areas. We propose an annotation schema to capture this information as a tuple of .