AI Scribe for Emergency Medicine: MDM & E&M Documentation

An AI scribe in the ED has to clear a higher bar than in primary care. The documentation can't just be a readable summary. It has to support your MDM-based E/M code selection, hold up under audit, and capture critical-care time precisely when it applies. And you have to get all of that done across a shift where you are managing several patients at once and getting interrupted roughly every six minutes.

An ambient scribe handles the capture part. The review and attestation stay with you. Here is what that looks like in a real ED shift.

Key takeaways

• A 2024 Yale study (JAMA Network Open) found ED attending physicians spend a median of 3.72 minutes per encounter on documentation. Modest on any single note, but the EHR time compounds across 15 to 20 patients in a shift.
• Since 2023, ED E/M codes 99281-99285 are selected on MDM alone. The note has to reflect the number and complexity of problems addressed, data reviewed, and management risk, not a time-counted exam.
• Critical-care time (CPT 99291/99292) requires an exact minute count and clear justification. Ambient capture at the bedside gives you the narrative to back that number.
• About two-thirds of ED attending time involves managing three or more patients simultaneously (observational workflow data). That parallel-patient model favors per-patient ambient sessions over push-to-talk dictation.

What changed in 2023 for ED E/M documentation

Before 2023, ED E/M coding could go through either the history and physical exam route or the MDM route. Starting in 2023, the AMA CPT guidelines for ED E/M codes 99281 through 99285 removed the history-and-physical counting method entirely. Code selection is now based solely on medical decision making.

MDM in the ED still rests on three familiar axes: the number and complexity of problems addressed, the amount and complexity of data reviewed and analyzed, and the risk of complications or patient management decisions. The highest two of those three determine the code level.

What this means for your notes: you do not need to count examination elements anymore, but you do need your note to clearly support the MDM complexity level you are selecting. A note that describes a straightforward complaint in generic terms will not support a 99284 or 99285 when an auditor is looking at it. A note that captures the acuity, the clinical reasoning, and the data you actually reviewed will.

An ambient scribe running during the encounter captures all of that in real time. You described the problem to the patient. You talked through the differentials with the resident. You reviewed the CT findings out loud. When the draft comes back two minutes after you close the session, the clinical reasoning you expressed during the visit is already in it. Not reconstructed from memory after five more patients.

How the MDM documentation fits together in a busy shift

Take a typical adult medicine ED shift: 15 to 20 patients, a mix of complaint types, some who bounce to observation, one or two critical cases. For most encounters the MDM documentation is the note. You need the history, the physical findings, what you ordered and why, and your impression and plan.

With push-to-talk dictation, the usual pattern is: see patient, return to the workstation between other patients, dictate from memory. In a fast department with interruptions coming in at about 10 times per hour, that memory gap is where documentation slips. The note ends up vaguer than the actual encounter was.

With ambient capture: open a session at the bedside, see the patient, close it when you leave. The note drafts while you are in the next room. By the time you have a free moment, the draft reflects what you actually said and did.

That is the practical difference. Not whether the scribe understands MDM theory. Whether the note it produces has enough specificity to support the code you are assigning. Your review of the draft is where you make that call.

If you are still working out which scribe features matter most for your setting, the 9-question evaluation scorecard at how-to-evaluate-ai-medical-scribe covers audio handling, coding honesty, export rights, and pricing transparency in detail.

Critical-care time: what the scribe gives you and what it doesn't

CPT 99291 covers the first 30 to 74 minutes of critical care; 99292 covers each additional 30-minute block. To bill either, you need an exact time documented, the time has to exceed 30 minutes, and the documentation needs to make the case for why that time was critical-care time.

Critical-care time includes: evaluating the patient, reviewing labs and imaging, speaking with EMS or transport teams, family discussions, consultant calls, documenting the visit, and retrieving outside records. It excludes time spent on separately billed procedures.

An ambient scribe does not count your time for you. What it does is give you a complete narrative: the EMS handoff discussion, the family conversation about goals of care, the attending-to-attending call to the ICU. That story exists when you sit down to document the time count. An auditor looking at a 99291 needs it. Without it, the minute number in the chart floats unsupported.

You still enter the exact count. You still attest. But you are doing that with a full draft already in front of you, not pulling a resus together from memory at the end of a twelve-hour shift.

The parallel-patient problem

Primary care has a sequential model: one patient at a time, a defined visit slot, a chart that waits. The ED does not work that way.

An observational study of ED attending workflow found that about two-thirds of attending time is spent managing three or more patients simultaneously. Interruptions come in at roughly 10 per hour; when a physician gets pulled away, the task they were on doesn't always get picked back up. That fragmentation is what pushes documentation out of the clinical window and into after-shift charting.

Ambient capture maps onto this model better than dictation does. You start a session when you walk into a room, you interact with the patient normally, and you close it when you leave. The capture is tied to that specific encounter. You do not need to reconstruct the visit from notes or memory later, and you do not need a quiet moment to dictate that may or may not come before end of shift.

The per-patient session structure also means you can run captures for different patients without mixing them up. Each encounter has its own draft. When you have a moment to review between patients, you work through the queue rather than rebuilding one long narrative from a fragmented shift.

What the note actually looks like after the scribe drafts it

AI Scribe by Patient Square is an ambient AI medical scribe that listens during the visit and hands back a structured SOAP note, ICD-10 suggestions, and a prescription draft — ready to review and sign about two minutes after the visit.

For an ED context, the SOAP format maps onto what an ED note needs: the subjective presenting complaint and history, the objective findings and data, the assessment including your differential reasoning, and the plan including disposition. ICD-10 suggestions give you a starting point for diagnosis coding. You confirm what is right, adjust what is not, and assign the final codes. The Rx draft covers medications you are ordering at discharge; yours to review before anything is finalized.

The note also includes enough clinical detail to support the MDM level because it captured your actual clinical reasoning during the visit, not a post-hoc summary. That is the difference from a template or a generic note.

To be direct about one thing: the scribe does not level your E/M code for you. It surfaces ICD-10 suggestions, not an MDM score or an automatic code assignment. The code selection, the critical-care time count, the attestation: all yours. What the scribe does is make the note that supports all of those things faster to produce and more complete when it arrives.

The audio question for an ED setting

Emergency departments are loud in a particular way: multiple conversations happening at once, other patients audible through curtains, staff talking behind you while you examine someone in front of you. That is the environment where some ambient tools quietly fall apart. Transcription accuracy drops, or the recording captures too much of the wrong room.

AI Scribe by Patient Square processes audio in memory and discards it the moment the note is drafted. No recording is kept. Nothing to retain, subpoena, or accidentally expose. In a shared clinical space where ambient sound includes half the department, that is not a minor implementation detail. It is what makes the tool viable in an ED at all.

Our security posture has data mapped to the HIPAA Security Rule, with BAAs available for every customer and SOC 2 Type II audit currently underway. If you want the technical detail on how we handle PHI in clinical environments, the security page covers it.

The cost math for an ED physician

An ED physician covering 15 to 20 patients a shift, working 12 shifts a month, is generating 180 to 240 encounters a month. At $89 per clinician per month on annual billing, you are paying less than 50 cents per encounter for the time-recovery. The math clears fast.

The harder-to-price return is the after-shift charting you stop doing at home. Per ongoing AMA tracking, EHR time continues bleeding into evenings for most physicians even as overall burnout rates inch down. The time compounds in the direction you probably care about most: finishing at the end of shift instead of at midnight.

The full pricing structure with no asterisks is on our pricing page.

Evaluating a scribe for ED work: the questions that are different

Most scribe evaluations center on note quality and price. That is correct, but the ED adds a couple of wrinkles worth pressing on.

The complexity test matters more here than in primary care. Ask a vendor for a trial on a real, representative case: an MDM 4 encounter, a chest pain workup, a resuscitation note. Grade the draft against what you would have written for that patient. If the note comes back generic when the encounter was complex, that is your answer before you spend a dime.

The audio question is also sharper in an ED context. When you ask a vendor what happens to the recording, the answer should be one sentence. "Processed in memory, discarded on note creation" is fine. "Retained for quality review" is not, full stop. You do not get to explain that one to a patient whose room shared walls with three others.

The trial design matters too. A seven-day trial on real ED shifts, with your actual patient mix, is the only credible data point. A scripted demo in a quiet exam room does not tell you anything about a Tuesday overnight.

On pricing: some scribes charge per note or per minute, which adds friction for shift-based practice. Know what you are paying per encounter at your volume before you commit. The full pricing structure for AI Scribe by Patient Square is on our pricing page, no asterisks.

Book a demo to run note quality against your specific visit types, then score the MDM specificity yourself on a real shift week.