Sleep medicine AI scribe 2026: STOP-Bang, polysomnography reports, CPAP titration, MSLT/MWT, and the outpatient + DME workflow

May 9, 2026 · 6 min read

Sleep medicine sits at the intersection of pulmonology, neurology, psychiatry, ENT, and dental sleep medicine, and the documentation reflects that breadth. A typical sleep clinic mixes outpatient evaluation visits, polysomnography (PSG) and home sleep apnea testing (HSAT) interpretation reports, CPAP / BiPAP titration narratives, MSLT and MWT for daytime sleepiness work-up, oral-appliance evaluation, and the DME prior-authorization substantiation that drives the practice's revenue floor. Generic ambient scribes capture conversation but underdeliver on the structured artifacts: PSG reports must conform to AASM scoring conventions; CPAP titration narratives must justify pressure settings to DME payors; MSLT interpretation must produce sleep-onset and REM-onset latencies in a payer-readable format.

The 2026 sleep-medicine-aware AI scribe stack handles five things general scribes do poorly: structured screening capture (STOP-Bang, Epworth, Insomnia Severity Index, RLS rating), AASM-aligned PSG and HSAT reports, CPAP / BiPAP titration narrative with payor-aware justification, MSLT / MWT interpretation, and the DME prior-auth substantiation that makes the difference between paid and denied claims.

Visit-type adaptation

Outpatient sleep medicine has five core visit types:

• Initial sleep evaluation — ROS-heavy across sleep, breathing, neuro, psych. STOP-Bang, ESS, ISI, RLS rating. Test ordering. 60-90 min.
• PSG / HSAT report interpretation — templated structure with AASM scoring (sleep stages, AHI / RDI by position and sleep stage, oxygen desaturation indices, arousal index). Often dictated separately from a face-to-face visit.
• CPAP / BiPAP setup and follow-up — pressure setting, mask interface, compliance review (usage hours, residual AHI, leak), troubleshooting. Compliance reporting drives Medicare DME continuation.
• MSLT / MWT interpretation — sleep-onset latency, REM-onset latency by nap, narcolepsy diagnostic criteria.
• Oral appliance / surgical referral — for CPAP intolerance or specific anatomy.

The scribe should detect visit type from context and apply the appropriate schema.

The sleep medicine system prompt

You are documenting an outpatient sleep medicine encounter or interpretation.

INPUT:
- Encounter audio transcript or PSG/HSAT raw scoring data
- Patient profile: prior sleep diagnoses, current PAP therapy and pressures,
  weight/BMI trend, comorbidities (HTN, AF, HF, T2DM, refractory hypertension)
- Prior screening scores (STOP-Bang, ESS, ISI)
- Compliance data (usage hours, residual AHI, leak) for PAP follow-up
- DME prior-auth requirements (Medicare LCD or commercial payor policy)

DETERMINE visit type, then apply schema:

For initial sleep evaluation:
1. Sleep complaint history (snoring, witnessed apneas, gasping, daytime sleepiness,
   sleep maintenance, sleep onset, restless legs, parasomnias)
2. Sleep schedule and habits (typical bedtime, wake time, naps, shift work)
3. Comorbid driver-of-OSA review (HTN, AF, HF, refractory HTN, stroke)
4. STOP-Bang score
5. Epworth Sleepiness Scale
6. Insomnia Severity Index if relevant
7. Mallampati class, neck circumference, BMI
8. Cardiopulmonary and ENT exam
9. Differential and test ordering rationale (PSG vs HSAT criteria)
10. Plan and follow-up timing

For PSG / HSAT interpretation:
1. Total sleep time, sleep efficiency, sleep latency, REM latency, WASO
2. Sleep stage percentages (N1, N2, N3, REM)
3. Arousal index (total + by cause)
4. Respiratory: AHI, RDI, ODI, AHI by position (supine vs non-supine), AHI by stage
5. Oxygen saturation: nadir, mean, % time below 88%
6. Periodic limb movement index
7. Cardiac: HR range, arrhythmia events
8. Severity classification (mild / moderate / severe OSA)
9. Treatment recommendation and rationale

For CPAP / BiPAP setup and follow-up:
1. Compliance data: average nightly usage, days of 4+ hours/night, % of days
   with use, residual AHI, leak
2. Pressure tolerance and mask fit
3. Symptomatic response (ESS change, witnessed apnea resolution, refreshment)
4. Plan: continue / pressure adjust / mask change / BiPAP escalation /
   alternative therapy
5. Medicare DME compliance documentation: 30-day, 90-day reporting where applicable

For MSLT / MWT:
1. Total sleep time prior night and PSG findings
2. Per-nap: sleep onset latency, REM onset latency, awakenings
3. Mean sleep latency
4. Sleep-onset REM periods (SOREMP) count
5. Narcolepsy diagnostic interpretation per ICSD-3 criteria

DME PRIOR-AUTH SUBSTANTIATION:
- Diagnosis ICD-10 (G47.33 OSA, G47.31 idiopathic, etc.)
- AHI / RDI from PSG or HSAT
- Symptoms qualifying (excessive daytime sleepiness, impaired cognition,
  cardiovascular comorbidity, refractory HTN)
- Conservative measures attempted (positional, weight, alcohol/sedative)
- Specific device, pressure, mask justification

Cite transcript or scored data. Use AASM standard vocabulary for sleep events.

STOP-Bang and ESS as the screening canon

STOP-Bang (snoring, tiredness, observed apnea, blood pressure, BMI, age, neck, gender) is the most widely used preoperative and clinic screening tool for OSA risk. Score 0-2 low risk; 3-4 intermediate; 5-8 high. Epworth Sleepiness Scale (ESS) quantifies daytime sleepiness across eight scenarios; 11+ suggests excessive sleepiness warranting work-up. A sleep scribe should always extract these scores when the items are covered in the visit, render them as standard score tables, and flag when items are missing so the clinician can complete the screen.

PSG and HSAT report templating

The PSG report is the highest-leverage artifact for a sleep practice: payors and referring clinicians read the report, not the visit note. AASM scoring rules govern arousal definitions, hypopnea definitions (1A vs 1B), and stage-by-stage breakdown. A scribe with PSG-aware templating produces a report shell from the raw scored data, leaves the interpretation prose to the clinician, and ensures every payor-required field is present (AHI, ODI, AHI by position, AHI by stage, %time SpO2 below 88%, sleep efficiency, etc.). HSAT reports are simpler but follow a similar template philosophy.

CPAP titration and DME compliance

Medicare and most commercial payors require documented compliance to continue PAP coverage past the 90-day trial: at least 4 hours per night on 70% of nights for any 30 consecutive days within the first 90 days. Compliance follow-up notes should explicitly capture the data downloaded from the device (or cloud platform) and the symptomatic response. Without that paragraph, DME claims get denied. A sleep-medicine-tuned scribe surfaces the required fields automatically.

The visit volume economics

A single sleep-medicine physician typically interprets 8-15 PSG / HSAT studies per day plus 8-12 outpatient visits, often in a hybrid practice. The mixed visit-and-interpretation pattern means audio-only ambient scribes only cover half the documentation work; the PSG / HSAT interpretation requires text-from-data templating. The DIY stack with both pipelines — audio transcription for visits, PSG-data-to-template prompt for studies — covers the practice end to end at $0.05/min on the audio side and near-zero marginal cost on the templated reports.

Vendor and DIY paths

Vendor ambient scribes capture conversation. They do not natively produce AASM-aligned PSG reports from scored data. Sleep-specific reporting tools (Compumedics ProFusion, Natus, Philips Sleepware) generate raw report skeletons but require physician overlay for interpretation prose. The DIY stack — LessRec Whisper API for visits + a sleep-tuned interpretation prompt that ingests scored data — produces both artifacts at variable cost.

BAA chain

Practice + EHR + sleep lab software (ProFusion, Natus, Sleepware) + transcription vendor + LLM vendor. DME prior-auth pathway adds the device manufacturer's portal (ResMed AirView, Philips Care Orchestrator) as a data flow.

When to start

Sleep medicine practices with substantial PSG / HSAT volume have the strongest case for a hybrid pipeline: the audio side covers visits, and the data-driven side templates reports. The DIY stack is the only configuration that handles both at the cost structure that matches the practice's mixed volume.