Recording ward rounds with both audio and video — on the premise that audio-only scribing cannot capture examination findings or procedures (BlackFrame.ai, NCT07649772)
AI tools that automatically document clinical conversations are being adopted rapidly in outpatient settings but have not been evaluated on hospital wards. Existing tools use audio only, which cannot capture physical examination findings, procedural observations or clinical safety behaviours — elements visible but not audible. This study evaluates an ambient audio-visual capture system in a real inpatient surgical ward.
Trial overview (primary data)
- StatusNot yet recruiting
- ConditionsClinical Documentation, Artificial Intelligence, Surgical Education, Medical Education
- InterventionsDEVICE: BlackFrame ambient audio-visual capture platform
- SponsorBlackFrame.ai
- Target enrollment60 participants
- Period2026-09-01 〜 2026-11-30
Key points
- A study evaluating an ambient audio-visual capture system, using both microphone and camera, in a real inpatient surgical ward. Planned enrollment is 60.
- It rests on the premise that audio-only tools cannot capture physical examination findings, procedural observations or clinical safety behaviours — what is visible but not audible.
- The platform generates a draft clinical note for a supervising clinician to countersign, and formative feedback for the trainee within 30 minutes.
- The record states expressly that no AI-generated text enters the patient record without explicit clinician review and sign-off.
- Of the 803 trials this site holds as of 2026-08-31, 26 have bodies touching documentation or scribing, against 33 touching chatbots.
- Physical examination and procedures are visible but inaudible, so an audio-only scribe leaves them out of the record.
1Recording what is visible but not audible
Automated clinical documentation has been built on the premise of turning conversation into text. What this study puts at its starting point is that the premise falls short on a ward. What was checked in a physical examination, how a procedure was carried out, what was done for safety — these are visible but leave no trace as sound.
Unlike an outpatient consulting room, a ward round proceeds at the bedside with hands on the patient. An audio-only record is missing part of what happened from the outset.
2Handling documentation and education through one mechanism
The system generates two things: a draft clinical note for a supervising clinician to review and countersign, and formative feedback for the trainee. The latter covers clinical communication, examination technique and documentation quality, delivered within 30 minutes. A record of the same scene becomes both the patient chart and teaching material.
A ward round has always been a teaching setting too, and whether parts the supervisor did not see can be reviewed bears on the quality of training.
3Stating that nothing enters the chart without a signature
What the record makes explicit is that no AI-generated text enters the patient record without explicit clinician review and sign-off. The generated draft stays a draft, and responsibility remains on the human side.
Of the 803 medical-AI clinical trials this site holds as of 2026-08-31, 26 have bodies touching documentation or scribing — second to the 33 touching chatbots, showing clinical AI reaching beyond diagnosis into administrative burden. How much documentation time was saved, and whether feedback improved performance, are not part of this record.
4What can be seen but not heard
Automated documentation of a consultation has been built on turning speech into text. What this study starts from is that the premise is insufficient on a ward.
The system generates two things: a draft clinical note for an attending to review and countersign, and formative feedback for residents delivered within 30 minutes, covering clinical communication, examination technique and note quality. The record states that AI-generated text never enters the patient record without explicit attending review and signature.
Why it matters
Automated documentation has rested on the premise of turning conversation into text, and where hands touch a patient that premise leaves records incomplete. Widening what counts as recordable changes the kinds of data handled and the design of consent and retention. Drawing the line at nothing entering the record without human review is a design worth referencing for keeping responsibility located.
FAQ
Why record video as well?
Does AI text go straight into the chart?
Was an effect confirmed?
Sources (primary)
Source: ClinicalTrials.gov (U.S. NIH/NLM, public domain). This site does not provide medical advice. Verify the latest and exact details with the official source. This site is not endorsed or certified by the NIH/NLM.
- ClinicalTrials.gov (study record, original)
- NCT ID: NCT07649772