TrialsgrambetaRCTs are beautiful

Lukac ambient AI scribes

2025

Ambient AI Scribes in Clinical Practice: A Randomized Trial

In 238 outpatient physicians across 14 specialties, an ambient AI scribe reduced writing time-in-note compared with usual care without a scribe.

Result

Primary endpoint: Change from baseline in log writing time-in-note

Time-in-note vs control: Nabla −9.5% (95% CI −17.2% to −1.8%; P=0.02); DAX −1.7% (95% CI −9.4% to +5.9%; P=0.66). Mini-Z improved with both (DAX +2.83; Nabla +2.69) and physician task load fell (DAX −39.9; Nabla −31.7). Scribes used in 33.5% (DAX) and 29.5% (Nabla) of visits.

Original publication
Subtopic
Ambient documentation
Design
Three-arm pragmatic RCT (covariate-constrained randomisation)
Relevance
Practice-defining: a trial the field needs to know.
Source
Lukac PJ, et al. NEJM AI. 2025;2(12).doi 10.1056/aioa2501000

Discussion & critique

The first randomised trial of ambient AI scribes, and a sobering one: the headline time saving was modest and reached significance for only one of two products, while scribes were switched on in under a third of visits. Burnout, task-load and exhaustion scores moved in the right direction for both, but these were secondary, survey-based and unblinded. Clinically significant inaccuracies were reported 'occasionally' for both scribes, so vigilance over note content remains the user's job.