Hawaii Pacific Health Medical Group piloted Abridge's ambient AI scribe with Epic, reducing note-writing time per appointment by 21% among high users, from 7.14 to 5.63 minutes. Time spent on notes per day also fell by 21%. The prospective study found no statistically significant change in burnout symptoms or patient satisfaction.
The project addressed the burden of clinical documentation during and outside office hours. Its evaluation compared provider perceptions with recorded EMR activity, rather than relying only on reported time savings. The paper distinguishes the relatively modest measured daily savings from the larger savings physicians perceived.
The medical group recruited 80 providers for onboarding in November 2024. Participants attended an instructional session and completed a prepilot survey before using the tool. After familiarisation, the pilot ran from December 2024 through February 2025; 79 providers completed it across 23 specialties, generating more than 25,600 notes. The group serves patients through hospitals and clinics across Hawaii.
Abridge was integrated with the Epic EMR. The researchers combined Epic Signal charting metrics, Abridge usage data, matched provider surveys and patient experience surveys. High users were defined as providers using the AI scribe for at least 60% of notes; 45 providers met this threshold. Objective time comparisons were limited to that group, whereas provider survey analyses included all participants.
The workflow required verbal patient consent documented in the encounter note. Physicians audited AI-generated notes before submitting them to the medical record. According to the AI company, recordings and transcripts were stored on secure HIPAA-compliant servers for 30 days before deletion. The research team deidentified data for publication.
Among high users, average daily note-writing time declined from 66.24 to 52.63 minutes. After-hours charting time declined numerically, but that change was not statistically significant. Provider-reported workload improved, while safety assessment relied on provider self-audits and was therefore subject to bias. The authors also noted that not all participants became high users, limiting statistical power.
Future evaluation priorities include larger cohorts, specialty comparisons, validated burnout measures, recording patient refusals and examining non-English encounters, where providers reported contextual and translation difficulties. The study identifies order queuing and service-level recommendations as planned integrations, not demonstrated benefits of this pilot.