Over the last several years, novel algorithmic techniques to process natural language powered by advances in machine learning have been developed. In numerous industries including healthcare, developers have demonstrated that these techniques can be used to perform certain user-specified tasks, which were previously not possible. Mutuo Health has recently developed a software model called AutoScribe that uses certain machine learning techniques to recognize patient-clinician dialogue and extract language features from the dialogues that are pertinent to a clinical note. Performance of AutoScribe is currently variable as the data that was used to train its models were the audio and transcript files only, without the medical note corresponding to the clinical visit. In this study, we want to determine whether it is feasible to adapt our model for the real-world clinical settings of primary care. We will optimize the AI models powering AutoScribe using the entire continuum of required data (audio, transcript and corresponding medical note) collected from primary care clinical setting.
This study aims to move Autoscribe from protocol to evaluation in a real world practice. If Autoscribe is optimized and found to be feasible within a diverse patient clinic setting, it may save time spent on notetaking, and increase patient-physician interaction satisfaction.