SHADOW Heidi Study
Research type
Research Study
Full title
SHADOW: Secondary-care Heidi Ambient Documentation Outcomes in Workflows: An assessor-blind non-inferiority/superiority evaluation of Heidi AVT documentation accuracy in secondary care
IRAS ID
370393
Contact name
Benjamin Austin
Contact email
Sponsor organisation
Heidi Health
Duration of Study in the UK
0 years, 5 months, 2 days
Research summary
When doctors see patients, they write notes about the consultation. These notes are important for patient safety, but research shows that up to 90% of clinical notes contain at least one error, usually information that was discussed but not recorded.
Heidi is an Ambient Scribe that listens to the doctor–patient conversation (with permission) and automatically creates a written summary of the consultation. This study will test whether the notes created by Heidi’s AI are as accurate as, or more accurate than, the notes doctors write themselves.
The study will take place across four hospital settings within NHS Greater Glasgow and Clyde: Emergency Department, Acute Medicine (Same Day Emergency Care), medical outpatient clinics, and surgical outpatient clinics. We aim to record up to 500 consultations (producing 1,000 notes — one AI-generated and one doctor-written for each consultation).
Importantly, the AI-generated note will NOT be used in patient care. The AI runs in ‘shadow mode’: it creates a note for research purposes only, and the doctor will not see it. Patient care continues exactly as normal.
After the consultations, trained reviewers (who do not know which note was written by the AI and which by the doctor) will compare both notes against a full transcript of the conversation. They will count any errors, things that were missed or incorrectly recorded, and assess how serious those errors could be.
The main risk to participants is the privacy risk from having their consultation audio-recorded. Audio recordings will be used only to create a written transcript and will then be deleted. All data will be stored securely and pseudonymised.
This study will provide the first large-scale, prospective evidence on the accuracy and safety of AI-generated clinical documentation compared to routine doctor notes in UK secondary care.
REC name
East of Scotland Research Ethics Service REC 1
REC reference
26/ES/0042
Date of REC Opinion
26 May 2026
REC opinion
Further Information Favourable Opinion