Why your hospital rates are what they are
Half the surgical work at a typical hospital does not cover its cost under Medicare. Nobody stops doing it. Somebody covers the difference, and if you buy commercial coverage, that somebody is you.
Stenting costs the hospital more than Medicare pays for it. Joint replacement roughly breaks even. Yet a joint replacement is half hardware by cost, and a stent case is only a fifth. So the device is not what decides it. The rate is.
Why this is your problem
A hospital that loses money on a procedure under Medicare has three options. Stop doing it, get cheaper at it, or charge someone else more.
We can rule out the first one. Hospitals with the worst Medicare margins on these procedures perform just as many as hospitals with the best. Nobody is walking away.
So the loss is being carried. A hospital that runs an operation below cost, keeps doing just as many of them, and stays open is meeting that shortfall from somewhere, and the only somewhere left is everyone who is not Medicare. That is what your rate is doing. Some of it buys care and some of it backfills a procedure the government underpays for.
Which makes the next question the useful one: what is actually driving the cost on the procedures you are backfilling, and is any of it moveable?
Every point is a device-led procedure. The two emphasised are the extremes of device content: joint replacement is the MOST device-intensive at 51% and sits near break-even; stenting is among the least at 20% and is furthest past it. Position on this chart tracks the horizontal axis, not device content.
Where the rate comes from
Medicare does not price a procedure by asking what it costs today. It looks at what hospitals reported spending on it, averages that across every hospital in the country, and pays that. The figures it uses are two to three years old by the time they take effect.
There is one more rule that matters. When one rate goes up, others must come down, because the total is fixed by law. So a rate is not really a price. It is a share of a pot, set by an average of everyone's history.
Now put a purchased device inside that. The rate moves slowly, backwards-looking, and shared. The device price moves whenever a manufacturer and a hospital sign a contract. The two have no reason to stay in line, and over time they do not.
That is the whole mechanism. A spinal fusion is paid $44,549 because fusions used to cost about that. A stent case is paid $12,119 because stent cases got faster and shorter, so the average came down. The stent itself did not get cheaper. The payment did.
Device content is MEASURED, not assumed: each construct is priced from the components a case consumes. The most device-intensive procedure here has among the better margins; the least device-intensive has the worst.
| DRG | Medicare pays | Costs | Device | Device % | Cost % |
|---|---|---|---|---|---|
| 454 Combined anterior/posterior fusion | 44,549 | 37,381 | 9,374 | 25.1 | 83.9 |
| 460 Spinal fusion except cervical | 25,514 | 25,372 | 9,374 | 36.9 | 99.4 |
| 321 Percutaneous cardiovascular, with MCC | 20,696 | 22,616 | 3,296 | 14.6 | 109.3 |
| 470 Major hip and knee replacement | 13,086 | 13,679 | 7,017 | 51.3 | 104.5 |
| 322 Percutaneous cardiovascular | 12,119 | 16,377 | 3,296 | 20.1 | 135.1 |
The rate is an average of what the procedure used to cost, everywhere. The device price is what one hospital agreed to last year. Nothing keeps them aligned.
Device-led procedures do run worse than medical admissions as a group. But within them device content does not explain the ordering, so read this as a fact about procedural work and its payment rates, not as evidence that devices cause the loss. Levels move ~24 points on the cost-to-charge basis; the ranking does not.
Buying more does not help
The obvious answer is scale. Get bigger, buy more, pay less. It is what group purchasing exists to do.
We checked it against what hospitals actually pay. A hospital buying $2M of devices a year and one buying $200M pay about the same. The correlation is 0.130, which is another way of saying there is no relationship.
That is the fifth explanation we have tried and failed to make work, after product mix, the hospital's size and owner, the manufacturer, and money paid to its surgeons. What a hospital pays for a device is not a function of how much it buys.
A spend-weighted average suggested large buyers pay more. The deciles refute it: that was a weighting artifact, not a relationship.
What you can actually do about it
Two things set the gap: what Medicare pays, and what the hospital pays for the device. You cannot touch the first. It is an average of what every hospital in the country used to spend, and it is set by statute.
The second is different. Device prices vary 2.6× between hospitals for the same components, and nothing about the hospital predicts where it lands — not size, owner, system, or scale. Between 12% and 24% of measured device spend sits above what better-buying hospitals pay. That is a modelled figure and it assumes those prices are reachable, which is only partly true.
For a purchaser that is the whole point. The device line is the one number in this chain that is large, variable, and derivable from files your hospitals already publish. You can put a figure on it before a renewal instead of after one, and it is the only part of the gap that anyone at the table can move.
You are already paying for the gap. The only part of it anyone can move is the part nobody can currently see.
Observed means read directly from a filing. Derived means calculated from filings by a stated method. Comparative means measured against a stated peer set. No finding is presented as more grounded than the weakest evidence behind it.
Device-led procedures run 90% to 119% of Medicare payment against 80% to 90% for medical admissions.
Each hospital's submitted charge for the whole claim x its own cost-to-charge ratio, against its observed Medicare payment, FY2024.
What would change it: A charge-weighted departmental ratio matching each claim's service mix, which is not derivable from public files.
Device share of cost does not predict margin (r = -0.284, wrong sign); the Medicare payment level does (r = -0.835).
Device constructs priced from measured components per procedure, against derived total cost and observed payment across 6 device-led DRGs, FY2024.
What would change it: More DRGs — six points is indicative, not robust — and multi-device cases, since the constructs price a single stent and a single-level fusion.
Purchasing scale does not predict what a hospital pays for devices; the correlation is 0.130 and the deciles are flat.
Device price index against annual implantable-device spend across 1,464 hospitals.
What would change it: Contracted volume rather than spend as the scale measure.
Between 12% and 24% of measured device spend sits above what better-buying hospitals pay.
Repricing each hospital's device spend at a median or upper-quartile price index, across $29.56B of measured spend.
What would change it: Evidence on how much of the gap is geography, contract timing and volume commitments that cannot be adopted.
A hospital running these procedures below cost without reducing volume is recovering the shortfall from payers other than Medicare.
Deduction from three measured findings: the median hospital is underwater on these procedures under Medicare, volume is flat across margin quintiles, and the institutions remain open.
What would change it: A like-for-like commercial rate, segmented by line of business, across many hospitals. A probe on 2026-08-04 found 4.5x Medicare at one payer and roughly parity at another.
- 1Cost is each hospital's own submitted charge for the whole claim multiplied by its own cost-to-charge ratio from HCRIS, the standard MedPAC approach, compared with the Medicare payment observed for that hospital and DRG.
- 2Procedures are grouped as device-led where an implanted device is the dominant purchased input: cardiac valve, defibrillator, pacemaker, percutaneous cardiovascular, spinal fusion, joint replacement.
- 3The device price index compares each hospital to the national median for the same component, part by part, so product mix is controlled by construction.
- 4The spend gap reprices each hospital's actual implantable-device cost at a median or upper-quartile index and is a counterfactual, not an observation.
- 1The device-share test rests on six DRGs. Six points is indicative, not robust.
- 2Device constructs price a single-stent PCI and a single-level fusion. Real multi-stent and multi-level cases carry more device content, which would raise the device share of exactly the procedures that already have the best margins, strengthening rather than weakening the result.
- 3Levels are sensitive to the cost-to-charge basis: a whole-hospital ratio and a surgical-department ratio move every group by roughly 24 points. The RANKING is stable; the decimals are not. Read the ordering.
- 4Fully-allocated cost carries overhead. A hospital deciding whether to take the next case reasons on direct cost, which is lower.
- 5Outpatient procedures are deliberately excluded. The only cost-to-charge ratio available for them is outside its validated range, and including them would have produced numbers we do not believe.
- 6Device content per DRG is inferred from procedure type, not measured line by line.
- 7The spend gap is modelled and assumes better-buyer prices are attainable by others, which is partly false.
- 8Medicare fee-for-service only. Commercial payment, which is where the loss is carried, is outside this analysis.
CMS Medicare Provider Utilization and Payment Data
Inpatient, by provider and service, FY2024 · 2024
Discharges, submitted charges and Medicare payments by hospital and DRG
CMS Healthcare Cost Report Information System (HCRIS)
Worksheet C Part I · Latest filed fiscal year per hospital
Cost-to-charge ratios, whole-hospital and departmental
CMS Hospital Price Transparency machine-readable files
Implantable device C-codes, 3,662 hospitals · Current published file per hospital
Component-level charges behind the device price index
Every figure in this report is reproduced by ops/sql/analysis/device-acquisition-cost-2026-08-03.sql. Where a number could not be measured, it is not claimed.