Impact of Unrestricted Artificial Intelligence Scribe Deployment in a Single Health System on Clinician Workload and Productivity

This is the authors' abstract. We don't add key points for this paper.

BACKGROUND: Ambient artificial intelligence (AI) medical scribes may reduce documentation burden and clinician burnout. Whether use intensity drives benefits remains uncertain. OBJECTIVE: To evaluate the association between AI scribe (Heidi) enrollment and documentation burden and whether changes differ by use intensity. DESIGN: Retrospective quality improvement study using an interrupted time series (ITS) indexed on each clinician's enrollment month adjusted for patient demographics and visit complexity. A secondary 3-group ITS compared enrolled nonusers, low users, and high users. SETTING: 14-hospital health system in Boston, Massachusetts. PARTICIPANTS: 902 outpatient clinicians enrolled in Heidi; 605 users and 297 enrolled nonusers. INTERVENTION: Heidi enrollment after system-wide implementation beginning January 2025. MEASUREMENTS: The primary outcome was note-active time. Secondary outcomes included same-day and after-hours note closures, note length, and work relative value units (wRVUs). RESULTS: 1 637 227 encounters had notes authored by enrolled clinicians between 1 July 2024 and 31 January 2026. Enrollment was associated with an immediate reduction in note-active time (-1.21 minutes [95% CI, -1.74 to -0.68 minutes]), with larger reductions among high users (-2.50 minutes [CI, -3.55 to -1.45 minutes]) than nonusers (difference, -2.20 minutes [CI, -3.46 to -0.95 minutes]) and low users (difference, -1.72 minutes [CI, -3.03 to -0.40 minutes]). High users had greater increases in note length than nonusers (difference, +170.90 characters [CI, 21.07 to 320.70 characters]) and low users (difference, +270.70 characters [CI, 131.04 to 410.36 characters]). No differences were observed in same-day closure, after-hours closure, or wRVUs. LIMITATION: Observational design with potential unmeasured confounding; inability to distinguish AI-scribed from clinician-authored notes postenrollment. CONCLUSION: Heidi enrollment was associated with reductions in documentation time and increases in note length, concentrated among high users. Enrollment was not associated with improvements in same-day, after-hours note closures, or wRVUs. PRIMARY FUNDING SOURCE: None.