Discovery and Visualization of New Information from Clinical Reports in the Electronic Health Record

Locating needed clinical information within electronic health record (EHR) systems can be difficult. EHRs contain a mixture of highly structured data, semi-structured templated data, and unstructured narrative as free text, not all of which are amenable to common search tools. Over time, the volume of information available within an electronic record has exploded, making it increasingly difficult for busy clinicians with limited time to quickly locate information needed for optimal patient care. Compounding the issues is the large amount of redundant information carried from one note to the next, with little clinical relevancy, and the difficulty in locating new information in the record. With little research about the underlying cognitive processes used by healthcare providers when they review patient data within an electronic record, user interfaces have not been optimized for digesting and utilizing clinical notes. This project sought to develop and validate an automated solution to detect new information in the EHR, with the goals of improving clinicians’ efficiency and decision making and decreasing their cognitive load.

The specific aims of the project were as follows:

• Refine computational methods to identify new information in clinical notes. 
• Assess the effect of visualizing new information in clinical notes. 
• Discover elements of a rationally designed EHR graphical user interface (GUI) to facilitate clinical document usage in practice. 

The project team developed an automated method for identifying relevant new information versus redundant information in EHR clinical notes using probabilistic language modeling. A visualization tool was then created that allowed providers to quickly identify information that was new in clinical notes. This tool was successfully implemented within a test EHR environment to complete proof-of-concept testing with providers. A series of usability testing was conducted around EHR use patterns with notes and EHR reading patterns by users in order to determine how physicians prefer to read progress notes and access other clinical data information in the EHR. Participants were asked to use think-aloud methods while performing tasks, and the data were analyzed to determine the amount of time required to assess the patient cases, perceived complexity of each case, and usability of the navigator.

The project team found several key patterns of EHR usage related to reading and creating notes. Users typically prefer a single note-writing style and do not deviate from their preferred template, while note-reading patterns relied heavy on the circumstances that initiated the task. Note-reading patterns were not always indicative of note-writing patterns and demonstrated that providers have different workflows and needs when reading a note as opposed to writing a note. Findings of the studies will be used to inform future design considerations of the GUI to improve usability and decrease clinician cognitive burden.