Home / Platform / Clinical Documentation

Module 01 · Structure-first ambient scribe

Give the exam room back to the patient.

Tap once, stop typing, sign on the way out. The note is structured, coded, and traced to the audio that justifies it.

app.rev.health/encounter — structured draft note with evidence links
rev.health clinical documentation screen with structured note sections

The signature mechanic

Ambient audio in. Coded, evidence-linked fields out.

The conversation becomes diarized utterances; the model proposes structured clinical facts; each fact keeps a link back to the exact 5-second window of audio that justifies it. Nothing reaches the chart until you accept it.

Ambient audio to structured-field write-back with evidence links A waveform of the visit conversation on the left, three diarized utterances in the middle, and three coded structured fields on the right (a Problem coded in SNOMED and ICD-10, an Assessment/Plan line, and an Order). Each output field is connected by an evidence link back to the utterance and its audio timestamp. AMBIENT AUDIO DIARIZED UTTERANCES STRUCTURED WRITE-BACK 11.4 min · es-US/en-US · AES-256 at rest · consent captured 02:1406:4809:31 PATIENT · 02:14 "My sugars have been running high all month." CLINICIAN · 06:48 "We'll start metformin 500 twice daily." CLINICIAN · 09:31 "Let's recheck an A1c in three months." PROBLEM · ADD Type 2 diabetes mellitus SNOMED 44054006 · ICD-10 E11.9 evidence → 02:14 · confirmed on accept ORDER · MEDICATION Metformin 500 mg PO BID RxNorm 861007 · queued to eRx evidence → 06:48 A/P · NOTE LINE Recheck HbA1c in 3 months LOINC 4548-4 · lab order evidence → 09:31 Model name + version stamped on every proposal (HTI-1 §170.315(b)(11) DSI transparency).

A bolt-on scribe would stop at the prose. rev.health's ScribeStructuredOutput rows each carry an evidence-utterance range, so click-to-play works from any signed note line back to the audio — and the write-back to the patient's global record is atomic with its source-org stamp.

The problem

Bolt-on scribes generate prose. Your chart needs structure.

Most ambient scribes stop at a draft note. The free text still has to be reconciled into the problem list, the medication list, and the order set by hand — the clinician saves time on typing and spends it on plumbing.

AI is imperfect — so we add structure. Every encounter becomes a versioned form: the clinician talks; the platform fills bounded, coded fields. Each visit teaches the form, and the forms evolve toward steadily more efficient and more accurate encounters.
The design principle behind the rev.health scribe

rev.health rejects the prose-only split. The scribe is not a peripheral — it is a first-class data source whose output writes directly into the chart's structured fields, with the audio transcript and timestamp captured as provenance on every fact.

The scribe proposes structured updates — new problems (SNOMED-CT + ICD-10 coded), medication changes, allergies, orders. You review, accept, and sign. The platform handles everything downstream: the codes, the charges, the claim.

Key capabilities

What ships in the documentation surface.

Pre-visit briefing

A one-screen chart summary before you walk in: last visit, active problems, meds, allergies, due preventive care, recent labs — plus an outside-records snapshot pulled via TEFCA. Built from the patient's global record, not just your practice's slice.

One-tap ambient capture

A single button opens the visit's versioned form and starts capture from the room iPad, in-room mic, or paired phone. Patient consent is captured before recording begins. Real-time transcription with speaker diarization.

Structured draft in 60 seconds

Within a minute of “End encounter,” the draft note is ready: HPI, ROS, PE, A/P. Click any sentence and play the 5-second audio clip that justifies it.

Structured write-back

The defining capability. The scribe proposes coded problem-list entries, medication changes, allergies, and orders. On accept, each becomes a global clinical fact with full source attribution — not free text trapped in a note.

SOAP or APSO — your call

First-class support for traditional S→O→A→P and assessment-first A→P→S→O layouts. Per-clinician toggle, persisted; the underlying data model is identical.

Real-time coding suggestions

As the note takes shape, ICD-10-CM, SNOMED-CT, CPT, and E/M level suggestions appear with MDM rationale — each linked to the audio evidence that supports it. See Coding & CDS →

Review queue & undo

Nothing reaches the global record until you sign. A 24-hour undo window covers write-backs; afterwards, retraction becomes a versioned correction with the original preserved in the audit trail.

Encounter templates

Annual wellness, acute, follow-up, telehealth, Medicare AWV — templates pre-load relevant note sections, screening prompts, and order-set candidates. Curated primary-care defaults ship in the box.

Bilingual capture

Spanish-English visits transcribe in both languages with diarization labeling speaker and language; the structured note generates in English with original-language utterances preserved for evidence review.

The data model

Eight objects turn a conversation into a defensible chart.

The split is deliberate: the act of producing a note (audio, transcript, proposals, review) is the practice's operational data; the clinical fact it produces belongs to the patient and lives in the global record with a source-org stamp.

Encounter & EncounterNote

The visit shell, opened at check-in and closed at “End encounter,” and its versioned SOAP/APSO note. Every correction is a new version — the original never leaves the audit trail.

ScribeSession

One capture session: consent flag and method recorded before a byte of audio is persisted, encrypted at rest, retained 365 days by default (configurable to 10 years), purged automatically while transcripts remain.

ScribeUtterance

The diarized transcript layer — one row per utterance with speaker label, millisecond offsets, BCP-47 language, and ASR confidence. The raw evidence the note is built from.

ScribeStructuredOutput

The bridge to the chart: each proposed fact (problem, med change, allergy, order, note line) references the utterance range that supports it and carries model name + version for DSI transparency.

ScribeReviewState

Per-proposal review — accept, accept-edited, reject, defer, undo. Nothing is promoted to a global fact without an accept; the reviewer and timestamp are recorded.

Problem & ProblemList (global)

The patient's longitudinal list — deliberately not org-partitioned. Each fact carries SourceOrganizationID, SNOMED-CT + ICD-10 codes, and FHIR Condition clinical/verification status.

The org boundary is crossed exactly once. Audio, utterances, and proposals all live behind the practice's partition-key isolation; the only point data becomes a global patient fact is the atomic write-back on accept — either the fact and its source-org stamp persist together, or neither does.

Functional & non-functional targets

What the implementation is bound to.

The pipeline, in requirements

  • Pre-visit briefing assembled from the patient's global Problem, Med, Allergy, Lab, and Imaging records plus TEFCA outside records
  • Consent captured before any audio is written to durable storage
  • Structured proposals produced within 60 seconds of session end
  • Click-to-play from any signed note sentence to its 5-second audio clip
  • Nothing promoted to a global fact without an explicit accept
  • 24-hour retraction window; afterwards a versioned correction
  • Every read of a note, problem, or transcript writes a read-access audit record

The numbers behind them

NFRTarget
End-encounter → draft note latency≤ 30s median · ≤ 60s p95
Word error rate (clinical eval set)≤ 6%
Hallucinated PE findings≤ 1%
Scribe capture availability99.9% monthly
Throughput (5 clinicians, 28 visits/day)≥ 140 sessions/day, queue p95 < 3 min
Offline audio buffering on capture outageup to 4 hours, no loss
SNOMED-CT US Edition ICD-10-CM RxNorm / LOINC FHIR Condition (US Core) HTI-1 DSI transparency TEFCA pre-visit pull 42 CFR Part 2 consent gating

Workflow

The 18-minute follow-up that ends in the room.

Open the pre-visit briefing

Last week's HbA1c and the home BP log the patient submitted through the portal are already on screen.

Tap “Start scribe,” confirm consent

Then spend the next 16 minutes talking. No laptop. No clipboard. Eye contact.

Tap “End encounter”

By the time you reach the workstation the structured draft is ready — HPI, ROS, PE, A/P, with an evidence link on every line.

Review the proposals

“Type 2 diabetes mellitus without complications” for the problem list, metformin 500 mg PO BID as an order — accept both with two taps.

Sign

The note is signed, the chart is updated, the charge is captured. Total visit: 18 minutes. Pajama time: zero.

Three years later

The audit-defensible note.

A payer audit lands on a level-4 E/M visit. The auditor asks for documentation of medical decision-making complexity. Your compliance officer pulls the signed note — every A/P line carries an evidence-link button. Clicking “discussed risk of progression to diabetic nephropathy” plays a 7-second clip of you explaining exactly that.

The auditor doesn't just see a note. They see the conversation that produced it.

Evidence-link coverage: 100%. Every A/P sentence in a signed note carries a transcript-timestamp evidence link — no exceptions. An evidence-link gap blocks sign-off.

Who benefits

Clinician

18–28 patients a day, eyes on the patient, note signed before the last patient of the day leaves. Owns clinical accuracy; the platform owns the plumbing.

MA / RN

Rooms the patient with the briefing on the iPad; vitals, med rec, and structured intake are half-done before the clinician walks in — with the MA's own contributions attributed in the audit trail.

Patient

Consents to recording at the start of the visit and reads the same signed note in the portal — the actual note, not a redacted summary.

Compliance officer

Audits consent capture, audio retention, access logs, and the evidence chain behind every note line — from one surface.

Performance targets

The numbers this module is built to hit.

MetricTarget
Charting time per visit (end-encounter to sign-off)≤ 90 seconds median · ≤ 4 minutes p95
After-hours pajama time≥ 60% reduction within 90 days of go-live
Draft note latency after “End encounter”≤ 60 seconds p95
Structured write-back acceptance without edit≥ 70%
Notes signed before the clinician leaves clinic≥ 80%
Hallucinated physical-exam findings≤ 1% on sampled audit sets
Evidence-link coverage on signed A/P sentences100% — gaps block sign-off

Zero pajama time is a feature. Ship it to your practice.

Join the design-partner cohort and put the structure-first scribe in your exam rooms.

Join the waitlist