Can the Administrative Loads of Physicians be Alleviated by AI-Facilitated Clinical Documentation?

Champions of AI-facilitated clinical documentation have suggested that the emergent technology may decrease the administrative loads of physicians, thereby reducing cognitive burden and forestalling burnout. Explorations of physicians' experiences with automated documentation are critical in evaluating these claims.

To evaluate physicians' experiences with DAX Copilot (DAXC), a generative AI-facilitated clinical documentation tool.

Semi-structured interviews were conducted in August and September of 2023 with physician-users of DAXC.

A purposive sample of 12 interviewees, selected from 116 primary care physicians, employed at a multi-site academic learning health system.

After completing all 12 interviews, three study personnel independently analyzed and coded the transcripts. Reconciliation sessions were then held to merge the three analyses into one summary, eliminating redundant codes, and grouping findings into themes.

For a majority of interviewees, DAXC reduced the amount of time spent documenting encounters, and alleviated anxieties of having to retain important clinical details until there was time to make notes. DAXC also allowed physicians to be more engaged during appointments, resulting in more personable provider-patient encounters. However, some physicians weighed these benefits against an uneasy feeling that interviewees might be asked to see more patients if DAXC was mandated. Physicians also noted that the tool would occasionally imagine or misgender patients, offer unsolicited and inappropriate diagnoses, and mistake critical details in transcription. The few physicians less enthusiastic about the generative technology portrayed themselves as creatures of habit who had cultivated long-standing workflows and particular notation practices that DAXC could neither improve upon nor reproduce.

According to physician interviewees, automated AI-driven clinical documentation has the potential to significantly reduce the administrative burden associated with particular types of provider-patient encounters. Addressing the growing pains of the incipient technology, identified here, may allow for a broader applicability for clinical practice.