How to Document Complex Medical Decision Making With AI

MDM is where the clinical thinking lives

Medical Decision Making (MDM) is the hardest part of a clinical note to document well. It's also the most consequential - for coding, legal protection and continuity of care. Since CMS revised E/M guidelines in 2021, MDM level primarily determines the code level for office visits.

And yet physicians consistently under-document their decision making. Not because the thinking didn't happen. Because translating a complex mental process into structured documentation takes time and cognitive effort that physicians don't have at 6 PM with fifteen charts still open.

The irony is that physicians discuss their reasoning extensively during patient encounters. They explain why they're ordering a test, why they're choosing one medication over another, why they're referring to a specialist. All that reasoning is verbalized. It just doesn't make it into the note.

AI scribes change this by capturing the clinical reasoning as it's spoken and incorporating it into the documentation.

The three MDM elements that drive coding

Under current CMS guidelines, MDM is evaluated across three elements. The highest two of three determine the overall MDM level.

Number and complexity of problems addressed:

Level	Examples
Straightforward	One self-limited problem (URI, minor rash)
Low	Two or more self-limited problems, or one stable chronic illness
Moderate	One or more chronic illnesses with mild exacerbation, or two stable chronic illnesses
High	One or more chronic illnesses with severe exacerbation, or acute illness posing threat to life

Amount and complexity of data reviewed and analyzed:

This includes labs, imaging, outside records, discussion of findings with other physicians and independent interpretation of tests. Each category of data adds to the complexity score.

Risk of complications, morbidity or mortality:

Treatment risks, drug-drug interactions, decisions about hospitalization, and choices between risky alternatives all factor into the risk assessment.

Why MDM is hard to document manually

The challenge with MDM documentation is that clinical reasoning is nonlinear. A physician doesn't think "first I'll consider the number of problems, then the data complexity, then the risk." They think about the patient as a whole and make decisions that implicitly account for all three elements.

When they sit down to document, they often default to brief, action-oriented statements:

"Continue current medications. Follow up in 3 months."

That plan tells you what the physician decided. It doesn't tell you why. It doesn't document the alternatives considered, the risks weighed or the data that informed the decision. From a coding perspective, that note supports a straightforward MDM level even if the actual clinical reasoning was moderate or high complexity.

How AI captures the reasoning you already verbalize

During a typical encounter for a patient with diabetes, hypertension and early-stage CKD, a physician might say:

"Your A1c came back at 8.2, which is up from 7.1 last time. Your kidney function has dropped a little too - GFR is 52 compared to 58 six months ago. I'm concerned that the diabetes and the kidney disease are connected. We have a few options. We could increase the metformin, but with your kidney function declining I want to be careful there. I think we should start an SGLT2 inhibitor - something like empagliflozin - because it helps with both the blood sugar and has kidney-protective effects. There's a small risk of urinary tract infections and ketoacidosis, which we'll watch for. I'd also like to repeat your labs in six weeks instead of three months given the changes."

That 30-second explanation contains:

• Three active diagnoses (diabetes, hypertension, CKD)
• Data reviewed (HbA1c trend, GFR trend)
• Assessment of disease interaction
• Alternative treatment considered and rejected with rationale
• New treatment selected with clinical justification
• Risk discussion (UTI, ketoacidosis)
• Modified follow-up plan based on clinical concern

An AI scribe captures all of this and structures it into the assessment and plan. The same physician writing from memory at end of day would likely produce a fraction of this detail.

Before and after: AI-documented MDM

Manual documentation:

A: T2DM, uncontrolled. CKD stage 3a. HTN. P: Start empagliflozin 10mg daily. Repeat BMP and A1c in 6 weeks. Continue lisinopril and amlodipine. Follow up 6 weeks.

AI-documented from the encounter:

A: Type 2 diabetes mellitus, uncontrolled (HbA1c 8.2%, increased from 7.1% six months ago). Chronic kidney disease stage 3a (eGFR 52, declined from 58). Concern for diabetic nephropathy given parallel decline in glycemic control and kidney function.

P: Initiating empagliflozin 10mg daily for dual glycemic and renoprotective benefit. Metformin dose increase considered but deferred given declining renal function (eGFR 52). Discussed risks of SGLT2 inhibitor including urinary tract infection and euglycemic ketoacidosis. Patient counseled on warning signs. Continue lisinopril 20mg and amlodipine 5mg for blood pressure management. Repeat BMP, A1c in 6 weeks (shortened interval due to declining trends). Follow up in 6 weeks to reassess.

Same encounter. Same clinical work. But the AI-documented version explicitly captures the complexity of the problems, the data reviewed, the alternatives considered and the risk discussion. This supports moderate-to-high complexity MDM, which is what the physician actually performed.

Practical tips for AI-assisted MDM documentation

To get the best MDM documentation from an AI scribe, adjust your verbal habits slightly:

• State your clinical reasoning aloud. If you're choosing between two treatments, say why you picked one over the other. The AI captures it
• Reference data explicitly. Instead of silently reviewing labs on screen, verbalize what you see: "Your A1c is 8.2, up from 7.1"
• Name the risks. When discussing treatment options, mention potential complications. This serves double duty - it documents risk assessment and fulfills informed consent requirements
• Quantify when possible. "Blood pressure is 10 points higher" is better than "blood pressure is elevated"

These aren't artificial changes to your practice. Theyre the same things you'd tell a medical student to do during rounds. The AI just needs to hear them to document them.

Transcribe Health captures the clinical reasoning behind your medical decisions and structures it into documentation. Your MDM is already happening in the conversation - the AI makes sure it shows up in the note.