Hospitals do not cover their cost on joint replacement
The hardware in a knee replacement costs a hospital $7,017, around 60% of what Medicare pays for the whole operation. On a fully-allocated basis the median hospital does not clear its cost, and hospitals differ by 2.6× in what they pay for implantable devices.
It is derivable from two federal filings: what a hospital charges for an item, and what a charged dollar costs it. Across 359,848 components priced at every hospital that publishes one, what stands out is not any single part. It is how far apart hospitals sit from each other, and how little of the operation's payment is left once the hardware is bought.
Why this is your number
When a hospital quotes a rate for a joint replacement, you have two ways to judge it. Against what other hospitals charge, or against what the operation costs the hospital. Only the first has ever been available to you.
This is the second. The hardware in a knee costs the median hospital $7,017, and hospitals differ by 2.6 times in what they pay for the same category of part. Those are costs the hospital filed itself, not estimates.
That changes what the negotiation is about. A rate can be set against what the thing costs instead of against what other buyers were charged, and the part that is arbitrary can be named and argued.
This part spreads <b>2.1×</b> from the cheapest decile to the dearest. It is one component, shown because it can be named down to the manufacturer and size. The <b>2.7×</b> in the headline is the median spread across all 5,918 brand-identified components, not this part.
What the spread is, and what it is not
It is tempting to read this as price discrimination on a component: the same baseplate, sold dearer to one buyer than another. The data does not support that reading, and it is worth being exact about why.
Decomposing 256,211 hospital-component observations, the raw spread on identifiable components is 2.86×. Almost all of it is a hospital-level effect worth 2.63×: a hospital dear on one component is dear on essentially all of them, by about the same factor. Strip that out and the item-to-item residual is only 1.31×.
So the finding is about hospitals, not parts. Some hospitals buy devices materially better than others across their whole basket. Which specific component you look at barely matters.
One thing this cannot settle. A uniform contract disadvantage and a residual markup effect that the department cost-to-charge ratio fails to remove produce the identical signature, and no public data separates them. What can be said is that the ratio compresses rather than creates the gap: posted charges alone spread 5.83×, and the correction brings that to 2.86×.
Nothing about the hospital explains it
The obvious explanations are that bigger hospitals buy better, that systems negotiate harder, or that the cheap ones are simply buying a cheaper product. Each was tested against the same data.
None of them survives. The spread sits inside systems, not between them.
Each explanation was tested independently against the same hospital-level data. The index compares every hospital to the national median for the same item, so product mix is controlled by construction.
| Explanation | What was tested | Result |
|---|---|---|
| Product mix | Items naming a size vs items that do not | 2.8× either way. Mix explains nothing. |
| Hospital identity | Ownership, system, scale, geography | Flat. Nonprofit 1.04, for-profit 1.02, HCA 0.98. |
| Manufacturer | Brand + part + size, matched via FDA GUDID | 2.7×. Attune 2.7, Legion 2.7, Reline 2.7. |
| Industry money | Device-maker dollars per affiliated clinician | r = −0.097 across 1,472 hospitals. |
Ten deciles of device-maker money, 1,472 hospitals, and the price index never moves. The first decile's 1.18 is hospitals with thin affiliation data, not a payments effect.
Why a component-level number matters
A few thousand dollars on a part sounds small beside the bill for an operation. It is not. Measured on the same hospitals, the hardware is most of what Medicare pays the facility for the whole procedure.
At the 90th percentile the hardware alone reaches 110% of the inpatient payment and 153% of the outpatient payment.
Where that leaves the margin
For the 455 hospitals where both sides can be measured, the hardware is 63.6% of what Medicare pays for the outpatient operation, and 51.8% of the inpatient payment across 271 hospitals.
At the top decile it exceeds the payment outright, reaching 110% inpatient and 153% outpatient. Those hospitals lose money on the hardware before a room is opened or anyone is paid.
Two limits. A device cost derived from cost reports is compared with a payment that is directly observed, so the ratio is only as grounded as the derivation. And the payment buckets are broader than a knee, so read these as the share for major joint work rather than for one operation.
Bars show the LOWER of two defensible bases. Using the whole-hospital cost-to-charge ratio, 51.4% of hospitals are above break-even inpatient and 54.8% outpatient. Using a ratio built only from the departments a surgical case consumes — operating room, recovery, anaesthesia, supplies, implantable devices — those become 80.5% and 83.7%. The correct answer lies between: a real claim consumes both surgical and routine departments.
Which makes the arbitrary bit expensive
That changes what the spread means. If joint replacement were comfortably profitable, a device cost varying this much would be a margin story. It is not comfortably profitable. On a fully-allocated basis the median hospital sits at or past break-even on both bases tested.
How far past depends on a methodological choice worth being open about. Costing the claim with the hospital's whole-hospital cost-to-charge ratio puts 51% of hospitals above break-even on the inpatient operation. Costing it with a ratio built only from the departments a surgical case actually consumes puts it at 81%. Both are defensible; the truthful answer is between them, because a real claim consumes surgical and routine departments alike. What does not move is the direction: on either basis the median hospital does not clear its fully-allocated cost.
But underwater on Medicare is not underwater. Ranking hospitals by margin, the worst group runs far past break-even and the best well under it, and they perform about the same number of procedures. Nobody is exiting. The Medicare loss is being carried, almost certainly by commercial payers, where the same operation pays a multiple of the Medicare rate.
Fully-allocated cost also carries overhead, and a hospital deciding whether to take the next case reasons on direct cost, which is lower. The honest statement is that the procedure does not cover its share of the building under Medicare, not that every case destroys value.
Bar shows p10 to p90; the dark rule is the median. A one-level lumbar fusion construct costs a hospital more than a total hip.
Spine is the larger category
Joint replacement gets the attention, but spine has more comparable components than hip and knee combined: 54,983. A single-level lumbar fusion construct runs a median of $9,374, against $7,017 for a full knee.
Implant markup widened <b>31%</b> over the decade, from 2.87× to 3.76×. But the hospital-wide markup widened <b>23%</b> over the same period, from 4.61× to 5.68×. Devices are somewhat steeper than the average, not a category apart. Device cost has also held at a steady <b>~8% of all hospital cost</b> since 2013, so this is not suppliers taking a growing share.
What to do with this
Every figure here comes from files your hospitals already publish and already file. That means the same arithmetic runs on your facilities, by name, before a negotiation rather than after one.
Three things follow for anyone buying care. Ask what the device costs, not what it is charged, because the gap is roughly fourfold and widening. Benchmark your highest-volume facilities against the national median for the same part, because the spread is within systems, not between them. And treat a hospital's position in that spread as a negotiable fact, since nothing about its size, owner or affiliations predicts it.
How this was measured
Hospitals have published their prices since 2021 and filed Medicare cost reports since 1966. The first says what a hospital charges for an item. The second says what a charged dollar actually costs it, in a line called Implantable Devices Charged to Patients. Multiply one by the other and you have what the hospital paid, by the same arithmetic Medicare uses to set its own rates.
Item descriptions turn out to be a shared vocabulary, so a 50mm shell at one hospital sets beside the same 50mm shell at another. Where the description also carries the brand, the match is to the product.
Running this on your own hospitals
Nothing here required private data. Both halves are public: the machine-readable price file each hospital posts under 45 CFR 180.50, and the Medicare cost report each files annually. The join is on the CCN, and the arithmetic is a multiplication.
The work is in the parts nobody enjoys. Item descriptions have to be normalised before a component at one hospital can be set beside the same component at another. The device cost centre has to be preferred over the general supply line, and hospitals reporting charge-as-cost excluded, or the ratio collapses to one. Components have to be reassembled into the construct a surgeon actually implants, because no single line is the implant.
The full query set that produces every figure here is committed at ops/sql/analysis/device-acquisition-cost-2026-08-03.sql. The same data is queryable through the Vlada MCP server and API, which is how our customers point it at their own facilities by name rather than at national medians.
The largest cost in a procedure half of hospitals already lose money on is the one they have the least visibility into.
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.
Hospitals differ by 2.6× in what they pay for implantable devices overall.
Median ratio of each hospital's derived component cost to the national median for the same component, across 256,211 observations at 284 hospitals. Decomposition attributes 2.63x to the hospital and 1.31x to item-specific variation.
What would change it: Evidence separating a uniform contract disadvantage from residual markup the department cost-to-charge ratio fails to remove. Both produce the same signature and no public data distinguishes them.
US hospitals reported $62.39B in implantable device costs against $234.51B in charges in FY2023.
Summed directly from filed Medicare cost reports, deduplicated to one report per hospital, fiscal year and cost centre.
What would change it: Nothing about the figure itself; it is a filed total. Its scope excludes hospitals that file no usable report.
Product mix, hospital identity, manufacturer and industry payments each failed to explain the spread.
Four independent tests against stated comparison sets: size-naming split, ownership/system/scale/geography cuts, brand-level SKUs via GUDID, and payment deciles across 1,472 hospitals.
What would change it: A confound none of the four tests captured, most plausibly purchase volume, contract vintage or GPO tier, none of which are public.
A single-level lumbar fusion construct costs a hospital a median of $9,374.
Component costs classified from item descriptions and summed into the construct, across 790 hospitals.
What would change it: A different component set for the construct, or evidence that the classifier systematically misassigns parts.
Industry payments to affiliated clinicians show no relationship with what a hospital pays for devices.
Pearson r = -0.097 across 1,472 hospitals, flat across all ten deciles of dollars per affiliated clinician.
What would change it: A sharper exposure measure. 96.8% of hospitals have a paid clinician, so this variable barely discriminates; absence of correlation here is not absence of influence.
The implant is 52% to 64% of what Medicare pays the facility for major joint work.
Device construct cost derived per hospital from cost reports, against that same hospital's observed Medicare payment for APC 5115 and DRG 470, 2024.
What would change it: A narrower payment bucket specific to the knee, or evidence that hospitals performing these procedures differ systematically from those publishing usable device lines.
On a fully-allocated basis the median hospital does not clear its cost on major joint replacement under Medicare; between 51% and 81% are above break-even depending on the cost-to-charge basis used.
Each hospital's submitted charge for the whole claim x a cost-to-charge ratio, against its observed Medicare payment, 2024. Two bases tested: whole-hospital (51.4% inpatient / 54.8% outpatient above break-even) and surgical-department (80.5% / 83.7%). Direction is stable across both; the point estimate is not.
What would change it: A charge-weighted blend of departmental ratios matching each claim's actual service mix, which is the correct instrument and is not derivable from the public files.
Hospitals with the worst Medicare margin on joints perform about as many as those with the best.
Margin quintiles from 2,211 hospitals, 2024: worst quintile at 150% of payment, best at 70%, median services 84 and 100 respectively.
What would change it: A longer panel showing entry and exit rather than a single year's cross-section.
- 1Acquisition cost for a chargemaster line is the hospital's posted gross charge multiplied by that hospital's own department cost-to-charge ratio from HCRIS Worksheet C Part I, cost centre 07200 (Implantable Devices Charged to Patients), falling back to 07100 where 07200 is absent, joined on CCN and using the latest filed fiscal year.
- 2Hospitals reporting charge-as-cost are excluded from the ratio, because for those filers cost equals charge by construction and would flatten the ratio to 1.0.
- 3A part is comparable when its normalised item description appears at 25 or more distinct hospitals. Spread is the 90th percentile of derived cost divided by the 10th across hospitals for that part, then the median of those spreads.
- 4The hospital price index compares each hospital to the national median for the same item, part by part, then takes the median of those ratios, so mix is controlled by construction.
- 5Brand identification matches brand names from the FDA GUDID device registry into the hospital item description. Only distinctive tokens are used; collision-prone words such as Anthem, Pinnacle, Journey and Gamma are excluded.
- 6National device cost and charges are summed from the cost reports directly, deduplicated to one filed report per hospital, fiscal year and cost centre.
- 1The device cost-to-charge ratio is a department average across every implantable device a hospital buys. It cannot remove hospital-specific markup on a particular component family, so part of the measured hospital-level difference may be residual markup rather than a difference in what was paid.
- 2The margin result is sensitive to which cost-to-charge ratio is applied. A whole-hospital ratio and a surgical-department ratio move the share of hospitals above break-even from 51% to 81% inpatient. Both are defensible and the correct instrument — a charge-weighted blend matching each claim's service mix — is not derivable from public files. The direction is stable across both bases; the point estimate should be read as a range.
- 3The device is compared with the facility PAYMENT, not with a facility cost built on the same basis, and the payment buckets are broader than one operation: APC 5115 is all Level 5 musculoskeletal work and DRG 470 covers hips as well as knees.
- 4An identical item description establishes the part and its size, not always the manufacturer. Where a brand token is present the match is to the product; where it is absent, two vendors' 50mm shells share a description. The brand-identified subset spreads 2.7× and the full set 2.8×, so this does not appear to drive the result.
- 5The cost-to-charge ratio is a departmental average across all implantable devices at that hospital, not a manufacturer invoice. It is a defensible cost estimate by Medicare's own method and should not be described as an invoice price.
- 6The industry payments test uses a many-to-many clinician-to-facility affiliation, so a surgeon's payments attribute to every hospital they are affiliated with. Payments are normalised per affiliated clinician, but attribution remains approximate.
- 7Procedure volumes, where cited, are Medicare fee-for-service only. Medicare Advantage covers 51.7% of eligibles and is absent from those files, as is all commercial volume. No national all-payer total is claimed here.
- 8Hospitals that do not publish implantable device lines, or that file no usable cost report, are outside the measured set entirely.
- 9Purchase volume, contract vintage and GPO tier are not public. They are the most plausible remaining explanation for the spread and this report cannot test them, so the finding is that four named explanations fail, not that no explanation exists.
- 10Margin is measured on FULLY-ALLOCATED cost, which carries overhead. A hospital deciding whether to run the next case reasons on direct cost, which is lower, so these figures describe whether the procedure covers its share of the building, not whether an additional case is worth doing.
CMS Hospital Price Transparency machine-readable files
3,662 hospitals publishing implantable device C-codes · Current published file per hospital, as held 2026-08-03
Posted gross charge and the item description that identifies the part
CMS Healthcare Cost Report Information System (HCRIS)
Worksheet C Part I, cost centre 07200 · Latest filed fiscal year per hospital; national totals FY2013 to FY2023
Department cost-to-charge ratio, and the national device cost and charge totals
FDA Global Unique Device Identification Database (GUDID)
Device brand and company names · Release 2026-06
Brand identification used to test the manufacturer explanation
CMS Open Payments, General Payment files
Device and medical-supply payments, PY2019 to PY2025 · Loaded 2026-08-03
Industry payments to clinicians, used to test the fourth explanation
CMS Physician Compare facility affiliation
Clinician NPI to facility CCN · As of 2026-05-29
Links a paid clinician to the hospitals whose prices were measured
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.