Evidently Workflow Study: A Randomized Trial of Clinician-Facing AI Summarization

Outpatient specialists spend substantial time reviewing records and documenting care, activities consistently associated with clinician burnout. This randomized trial tests whether the Evidently clinician-facing AI summarization tool, embedded in Epic, reduces that burden in routine specialty practice.

Design. A waitlist-controlled, parallel two-arm randomized trial at UNC Health. Physicians randomized to immediate access received the Evidently tool and brief training at the start of an eight-week period; the delayed-access control arm continued usual chart review and received the tool after follow-up. Allocation was 1:1 by covariate-constrained randomization, balanced on baseline burnout, chart-review time, weekly clinic sessions, and specialty. Enrollment was 128 outpatient attending physicians (64 per arm) across the UNC Faculty Practice and UNC Medical Group, spanning cardiology, pulmonology, gastroenterology, neurology, oncology, surgery, and anesthesiology.

Outcomes. The primary outcome was the change in clinician task load over eight weeks, a Provider Task Load composite of mental demand, physical demand, time pressure, and effort, analyzed by intention-to-treat with adjustment for the randomization-balance covariates. Secondary outcomes covered burnout, chart-review time during and outside scheduled hours, per-new-patient preparation time, worry about missed chart information, and preparedness at the start of a visit. Supplementary analyses drew on objective EHR audit-log usage (Epic Signal) and OMOP-based measures of patient complexity.

The trial (NCT07498582) was sponsored by the University of North Carolina at Chapel Hill, with Spencer Dorn as principal investigator; I served as co-investigator and lead statistician, leading the randomization and statistical analysis. Primary data collection completed in June 2026; results are in preparation.