Ambient Documentation

What it is

Ambient documentation systems record the clinician-patient conversation, transcribe it, and generate a structured note that the clinician edits and signs. Most run partly on-device and partly in the cloud, with a shift toward local processing for privacy.

The pipeline

1. Capture — microphone array in the room or on a badge.
2. Speech-to-text — usually a domain-tuned ASR model.
3. Speaker diarization — patient vs. clinician vs. family.
4. Note generation — an LLM produces the SOAP/HPI structure.
5. Coding assist — ICD-10 / CPT / E&M level suggestions.
6. Clinician review, edit, sign — with edits fed back for improvement.
7. Post to EHR — via FHIR DocumentReference or an HL7 v2 MDM message.

What the evidence shows

Peer-reviewed evaluations from 2023–2025 (Kaiser Permanente, Stanford, The Permanente Medical Group, UNC) report:

• 20–60% reduction in after-hours documentation time.
• Modest to significant reductions in self-reported burnout.
• Modest improvements in patient-reported communication.
• Persistent risk of hallucinated facts and dropped negatives.

Governance

• BAA with the vendor, always.
• Documented retention for audio and transcripts (many programs delete audio within 24 hours).
• Patient notice and opt-out, ideally verbally at each encounter.
• Ongoing quality audits — sample notes for accuracy, drift, and hallucination.
• Alignment with FDA GMLP and ONC HTI-1 transparency requirements for predictive DSI.