Abstract
A growing body of work uses Natural Language Processing (NLP) methods toautomatically generate medical notes from audio recordings of doctor-patientconsultations. However, there are very few studies on how such systems could beused in clinical practice, how clinicians would adjust to using them, or howsystem design should be influenced by such considerations. In this paper, wepresent three rounds of user studies, carried out in the context of developinga medical note generation system. We present, analyse and discuss theparticipating clinicians' impressions and views of how the system ought to beadapted to be of value to them. Next, we describe a three-week test run of thesystem in a live telehealth clinical practice, major findings from whichinclude (i) the emergence of five different note-taking behaviours; (ii) theimportance of the system generating notes in real time during the consultation;and (iii) the identification of a number of clinical use cases that could provechallenging for automatic note generation systems.