Ten companies got a closed-door demo day with FDA and CMS, and part of the agenda was how Medicare should pay for clinical AI. This is the thing they were talking about: every price Medicare pays for an inpatient stay, all of them, in one field.
On July 8, officials from the FDA and CMS hosted what an internal agenda called a "clinical AI demo day" at FDA's White Oak headquarters. Ten companies presented. The meeting was never publicly announced; STAT reviewed the agenda and published it on August 5.
No health system. No specialty society. No nursing organization. No FQHC network. No open-source project.
The guest list got the attention. The agenda is the story: regulators were not only asking whether these tools are safe, they were asking how Medicare should pay for them. That second question is the one that decides what actually gets built, because clearance only decides what is legal to sell.
So here is what "how Medicare pays" looks like when you put all of it on one screen.
It plays on load. Each dot is one MS-DRG — one of the fixed prices Medicare pays for an inpatient stay, no matter what the hospital spent. Scrub back and forth, or replay.
The animation ends on a $25,000 technology. Move it yourself. The line is what the hospital has to absorb per case — the technology's cost, minus any new technology add-on payment, which pays 65% of the technology's cost when a product qualifies.
With no add-on payment, a $25,000 technology exceeds the entire average Medicare payment in 620 of 773 DRGs — 88% of every Medicare fee-for-service inpatient discharge in the country. With the add-on at 65%, that falls to 220 DRGs and 32% of volume.
That difference is the entire reason NTAP exists, and it is what CMS just made harder to get. The FY 2027 inpatient rule published August 4 repealed the alternative pathway that let an FDA Breakthrough Device designation substitute for proving substantial clinical improvement. Every applicant now demonstrates newness, cost, and improvement independently — roughly 12 to 18 more months of comparative-effectiveness evidence, during which the hospital eats the red dot.
Go to regulations.gov, find one open rule that touches what you build, and read only the “Comments Requested” section — agencies list the exact questions they want answered. Answer one, in a paragraph, from your own practice. A specific comment from a named practicing clinician is rare enough that agencies quote it in the final rule preamble. Every price on the ladder above was set through that process. It is slow, unglamorous, and it is the only room that is actually open.
Scroll the field again. There are 773 rungs on this ladder, and not one of them is an autonomous system evaluated the patient. Several of the companies in that room are building products with no clinician performing the service, which means no evaluation-and-management code fits and no DRG contains them. Somebody has to invent the category, and whoever writes the definition of "a unit of service when no clinician performed it" — per encounter? per decision? per avoided visit? — sets the business model for everything that follows.
That definition was being discussed on a Wednesday in July in Silver Spring, by ten companies and two agencies, with no public notice and no minutes.
These are payments, not costs. Medicare's file records what it paid and what hospitals charged. It contains no cost column. NTAP is calculated off the hospital's actual case cost, so treat the red dots as a ceiling test — can the payment even cover this technology on its own — not as a margin calculation. A real case has other costs, which makes it worse, not better.
Fee-for-service only. Original Medicare Part A at IPPS hospitals. No Medicare Advantage, which now covers roughly half of Medicare enrollees and sets its own terms; no Medicaid, commercial, uninsured, critical access or long-term care hospitals. 6.93 million discharges is not US inpatient volume.
The long tail is thin. 263 of the 773 DRGs saw fewer than 1,000 discharges nationally in the year and together account for 1.6% of volume; 36 saw fewer than 100. CMS suppresses cells under 11 discharges entirely. A dot is a row, not a population — counting rows is how “we address 700 DRGs” gets said out loud.
Averages are national. Wage index, DSH, IME and outlier payments move any single stay well off the mean. Vermont averages $9,357 per discharge across all DRGs; Maryland averages $21,790 and is on a global-budget waiver that makes the comparison shaky in the first place.
The vintage is a file date. The performance_year column is null for every row in this table; 2023-12-31 is the publication vintage used as the year. DRG definitions also change annually, so a code number does not mean the same thing across years.
65% is the current rate, not a constant. The marginal NTAP rate was 50% before FY 2020 and is 75% for qualified infectious disease products. If you have seen “up to 50%” quoted, that is the old rule.
| Cut | DRGs | Discharges | Share of volume |
|---|---|---|---|
| Average payment under $10,000 | 266 | 2,943,399 | 42.5% |
| Average payment under $25,000 | 620 | 6,093,840 | 88.0% |
| Average payment under $50,000 | 729 | 6,761,792 | 97.6% |
| Fewer than 1,000 discharges/yr | 263 | 113,157 | 1.6% |
| All DRGs | 773 | 6,926,093 | 100% |
Median DRG payment $12,714 · volume-weighted average $15,033 · cheapest DRG 310 (cardiac arrhythmia without CC/MCC) at $3,356 · most expensive DRG 018 (CAR‑T cell immunotherapy) at $416,656 · total Medicare payments $104.1 billion.