If you have ever finished a shift wondering whether your charts will hold up to a payer audit, the answer almost always comes down to one section: Medical Decision Making.
MDM is the dominant input into E/M leveling under the 2023 revisions and remains the area where emergency physicians lose the most billable complexity to incomplete documentation. The mechanics are straightforward once they're internalized, but most physicians were trained on older E/M frameworks and have never had a clean reference for what the current rules actually require.
This is that reference.
What MDM is — and what it isn't
MDM is the documented record of your clinical reasoning during an encounter. Under the 2021 outpatient and 2023 ED-applicable revisions, it is the primary determinant of E/M level for most visits. (Time can also be used as the determinant in many settings, but ED visits remain MDM-driven.)
Three pillars determine MDM complexity:
1. Number and Complexity of Problems Addressed. What was the differential? Which problems were ruled in or out? How acute, severe, or undifferentiated were they? 2. Amount and Complexity of Data Reviewed. Labs ordered and reviewed, imaging interpreted, prior records pulled in, independent visualization performed, conversations with consultants. 3. Risk of Complications and/or Morbidity from Patient Management. What was the risk of the management decisions made — including the decision to admit, discharge, prescribe controlled substances, or pursue invasive procedures.
The level of MDM (Straightforward, Low, Moderate, High) is determined by the highest two of these three pillars. This is the math that drives 99281 → 99285 ED levels.
For the official structure, the CMS E/M guidelines and the American Medical Association CPT documentation are the primary references. ACEP also publishes specialty-specific guidance at acep.org.
The most common MDM under-documentation patterns
In our reading of audit findings and post-hoc reviews of charts, four patterns recur.
Pattern 1: Documenting the diagnosis but not the differential. "Costochondritis" is a diagnosis. "Chest pain in a 34-year-old, ruled out ACS based on negative troponins ×2 and normal ECG, with PE excluded by Wells score and d-dimer, ultimately attributed to costochondritis based on reproducible point tenderness" is MDM. The first earns nothing. The second documents the actual cognitive work.
Pattern 2: Listing tests ordered but not their interpretation. Ordering a CT abdomen/pelvis with contrast adds complexity to the Data pillar. Reviewing it and acting on the findings is what makes it count. Notes that say "CT abdomen ordered" without "CT showed no acute findings; appendicitis ruled out" are leaving complexity on the table.
Pattern 3: Underplaying risk. Risk in the MDM sense is the risk of complications from the management decisions, not just the underlying disease. Discharging a patient on opioids, deciding to not admit a borderline case, pursuing a moderately invasive procedure, prescribing a high-risk drug — these are all moderate-to-high risk management decisions and should be documented as such.
Pattern 4: Missing the discussion. A 5-minute conversation with cardiology about whether to admit a borderline troponin patient counts toward Data complexity. So does an independent historian — speaking with the family member of an obtunded patient. These conversations are routine in the ED and are routinely missed in documentation.
How to structure MDM that holds up
The structure that works in practice — and is becoming standard in AI-generated ED notes — is a short paragraph for each of the three pillars, written in the voice of the physician, capturing the actual reasoning of the encounter. Not a checklist. Not boilerplate. Real reasoning.
A working ED MDM section under this approach is typically 150–300 words. It is not a full re-narration of the case (the HPI handles that). It is the physician's clinical reasoning, structured by the three pillars.
A worked example
For a 58-year-old with chest pain, ruled out ACS, discharged with cardiology follow-up, the MDM might look like:
Problems Addressed: Acute chest pain of undifferentiated etiology in a patient with cardiac risk factors. Concern for acute coronary syndrome was high on presentation, given age, hypertension, and smoking history. Pulmonary embolism considered given pleuritic component but lower pre-test probability. Aortic dissection considered and clinically excluded based on equal upper-extremity blood pressures and normal mediastinal silhouette on CXR.
>
Data Reviewed: Serial troponins (0h and 3h) negative. ECG without ischemic changes on serial recordings. CXR without acute pathology. D-dimer obtained and within range, supporting low PE probability. Outside records reviewed — no prior cardiac workup. Brief discussion with cardiology fellow confirming outpatient stress testing is appropriate disposition.
>
Risk: Moderate. Decision to discharge home from the ED with arranged outpatient cardiology follow-up was made after shared decision-making with the patient regarding risk-benefit of inpatient versus expedited outpatient workup. Return precautions reviewed in detail and documented.
This is moderate MDM for the Problems pillar (one undifferentiated chronic illness with significant risk of progression), moderate-to-extensive for Data (independent interpretation of tests, external records, consultant discussion), and moderate Risk (discharge decision in a patient with workup considerations). Two of three pillars at moderate maps to Moderate MDM, supporting a 99284.
The audit defensibility test
The simplest test for whether your MDM will hold up: if a payer auditor reads only your MDM section, would they understand what you were thinking and why you made the decisions you made? If yes, the chart is defensible. If they would have to reconstruct your reasoning from fragments scattered across HPI, exam, and orders, the chart is at risk.
The cleanest MDM sections read like the physician's actual clinical thought process — because that's what they're documenting. The work isn't to add more words; it's to capture the reasoning that already happened during the encounter.
Where AI is changing the equation
Modern ambient AI documentation tools generate first-pass MDM sections by structuring what the physician said during the encounter into the three-pillar framework. The clinical judgment remains the physician's; the structure becomes automatic. This addresses one of the longstanding gaps in ED documentation: physicians do the reasoning in real time, but capturing it in audit-ready prose at the end of a 12-hour shift is where the wheels come off.
Whichever workflow gets you there, the underlying rules are the same. Document the differential. Document the data, including its interpretation. Document the risk of the decisions you made. Two of three at the right level supports the level you bill.
