An ambient scribe vendor just shipped a tool that reads the bill — comparing the coded diagnoses and the assigned DRG against what the captured encounter supports, before the claim drops. Which makes this the number to look at: in 43 Medicare DRG families, the only difference between two payment codes is whether a major complication appears in the documentation.
Medicare pays an inpatient stay by MS-DRG, and for a large part of the list the severity split is not a clinical judgment. It is a lookup. CMS publishes the major complication and comorbidity list annually; either a documented condition is on it or it isn’t. Same surgeon, same implant, same three days — one word apart in the record, two different prices.
Ambient capture has been sold for two years on minutes saved. A tool that reads the captured encounter against the coded claim is not selling minutes. It is sitting on top of this list.
Minutes saved is a satisfaction metric. A DRG that moved because the machine heard something the physician never wrote down is a payment-integrity question.
Each line below is one DRG family with a clean two-way severity split — identical description, one ending “with MCC,” one ending “without MCC.” The left end of the line is the average Medicare payment on the without code. The right end is the with code. The filled dot is where the actual population landed: its position along the line is the share of cases coded to the upper tier.
These are observed average payments, not a coding experiment. A case coded with a major complication is, on average, genuinely sicker: it stays longer, consumes more, and is likelier to trip outlier payments. The line length is therefore the full observed spread between the two populations, not the marginal dollars a single documented word adds.
The marginal number is smaller and more knowable, because CMS publishes it. Today’s newsletter works the joint-replacement case: at the FY2026 relative weights — 1.9289 without an MCC against 3.0332 with one — the 1.1043-weight difference prices out near $7,500 per case against the national standardized amount. In this 2023 claims file the same pair shows an observed gap of $14,626, nearly double. Both numbers are correct; they measure different things, and a vendor pitch will quote whichever is larger.
MS-DRG 469 is not a clean pair. Its full title sweeps in total ankle replacement alongside the with-MCC cases, so 469/470 fails the “identical description, one word apart” test the 43 families above pass. It is also the thinnest cell in the file:
| Code | Discharges | Hospitals reporting | Avg Medicare payment |
|---|---|---|---|
| 470 — without MCC | 71,939 | 1,311 | $13,696 |
| 469 — with MCC or total ankle | 1,134 | 66 | $28,322 |
Sixty-six hospitals. CMS suppresses any provider-DRG cell under 11 discharges, so the 469 column is a self-selected slice of mostly large centers — which is another reason the $14,626 is not the marginal price of a word.
Nobody in this picture is coding fraudulently. That is the problem. A tool that reliably surfaces documentable severity raises severity on average, across every case, with no individual decision anyone could point to. The honest version and the bad version produce the same chart.
Across these 43 families, cases landing on the upper tier carried roughly $6.7 billion more in Medicare payments in 2023 than the same count of cases would have on the lower code. Move national capture a few points in either direction and that figure moves by hundreds of millions — without a single claim anyone would call wrong.
If your product touches documented severity, build the attestation artifact before the feature: which human reviewed which machine-surfaced finding, when, and what they changed. Not because anyone is asking. Because the auditor who asks in 2029 will be asking about 2026.
Next: the same question, one hospital at a time → — every hospital’s severity-capture rate in the three biggest MCC families, and what happens when you filter out the small ones.