Ambient Audio-Visual Capture for Clinical Documentation and Assessment

Overview

AI-powered tools that automatically document clinical conversations are being adopted rapidly in outpatient settings but have not been evaluated in hospital wards. Existing tools use audio recording only, which cannot capture physical examination findings, procedural observations, or clinical safety behaviours - elements of a ward round that are visible but not audible. This study evaluates an ambient audio-visual (AV) capture system - BlackFrame - that uses both microphone and camera to generate accurate clinical documentation and structured educational feedback in a real inpatient surgical ward setting. Medical students and doctors in training participate in supervised ward round encounters with consenting adult inpatients. The BlackFrame AI platform generates: (a) a structured draft clinical note for the supervising clinician to review and countersign before any use in the patient record; and (b) formative feedback for the trainee, delivered within 30 minutes, covering clinical communication, examination technique, and documentation quality. The study measures whether AI-generated feedback improves trainee clinical performance over a placement, how much documentation time is saved, and whether the system is acceptable to patients and clinicians. No AI-generated text enters the patient record without explicit clinician review and sign-off. All participation is voluntary.

BACKGROUND Ambient AI scribes have achieved rapid uptake in outpatient and community settings but all published evaluations use audio-only capture. The inpatient ward round is a multimodal clinical event comprising verbal exchange, physical examination, procedural assessment, and non-verbal observation. Audio-only systems are structurally incapable of capturing observable clinical elements, representing construct under-representation under the Messick validity framework. No published study has evaluated ambient audio-visual capture in a real inpatient setting, nor measured the educational impact of AI-generated formative feedback on ward rounds. STUDY DESIGN Mixed-methods feasibility and educational impact study. Surgical ward round at Yeovil District Hospital as the primary study context. Up to three ambient AV capture devices deployed simultaneously in separate side rooms on each study day. Ward rounds proceed sequentially through each room, allowing up to three consented encounters per study day. PARTICIPANTS Trainee participants: medical students (Year 3-5) and doctors in training (FY1 through registrar/ST grade) undertaking supervised clinical activities at the study site. Patient participants: adult inpatients (age 18 or over) able to provide informed consent, admitted under the surgical team, clinically stable at the time of approach. TARGET SAMPLE: 60-80 consented encounters across 20-30 trainee participants and up to 80 patient participants. INTERVENTION On each study day, eligible patients in up to three side rooms are consented before ward rounds begin. A BlackFrame ambient AV capture device is positioned visibly in each consented patient's room prior to the ward round, with clear patient-facing signage. Devices operate autonomously once positioned and do not require operator presence during the encounter. The surgical ward round proceeds sequentially through each side room. After each encounter the AI platform produces: (a) a structured draft clinical note for supervising clinician review and countersignature before any use in the patient record; (b) a formative feedback report for the trainee covering clinical communication, examination technique, and documentation quality, delivered within 30 minutes. OUTCOMES Primary: (1) Change in trainee assessment scores from baseline to end-of-placement; (2) documentation time saved with versus without AI s