DAX Copilot: ambient AI scribe may help reduce surgical resident clinical documentation burden.

introductionThe burden of clinical documentation is high, especially for surgical residents who must balance a high clinical and operative workload with accurate and sufficient documentation. This can lead to resident burnout and worse patient care and satisfaction. DAX-Copilot, an ambient-AI tool that can "listen" to a clinical interaction and draft a clinical note, may help alleviate this documentation burden. In this study, we critically assess the accuracy, usefulness, and readiness of DAX-Copilot for use by surgery residents in an inpatient setting.

methodsDAX Copilot, a commercially available AI-powered scribe that conforms to Health Insurance Portability and Accountability Act (HIPAA) standards was tested in 25 simulated inpatient scenarios. We simulated these patient interactions with multiple voices provided by the authors in a highly representative setting. Three authors used a Modified Patient Documentation Quality Instrument (PDQI-9) Assessment Tool to score the generated notes.

resultsOverall, the authors were impressed by DAX Copilot's ability to convert plain speech to accurate and consistent notes. The authors also noted that the AI was exceptionally easy to use and was able to recognize dissatisfied patients and describe their complaints in a professional manner. On average (SD), the generated notes scored 46.91/50 (1.35). More specifically, the notes scored highly in the following domains: accuracy, thoroughness, comprehensibility, succinctness, synthesis, and internal consistency, and were free from hallucination and bias. We scored the tool a 4/5 (0.5) for organization and usefulness because the default note still requires basic editing to fit the style of surgical notes.

conclusionAmbient AI is poised to revolutionize clinical documentation, with the potential to reduce its burden on surgical residents. More work is required to assess this tool in additional clinically relevant patient encounters. Additionally, the authors would like to explore methods to promote institutional proficiency in using this tool and further optimize its output.