Two public files put the median hip implant construct cost at $6,705 across 319 hospitals

Mark Cuban has been arguing that joint-implant pricing is deliberately hidden, and that prices tend to fall 30 to 40 percent once somebody publishes them. Dr. Jon Slotkin answered from the contract side with actual numbers: a hip construct runs roughly $5,000 to $6,000 per implant, and he has seen the same case span $1,800 to $12,600. Then Vlada Health published a method card on August 2, 2026, showing that you can get at that number without anybody’s contract at all: public files, 236 hospitals, four states. Florida $5,401, Texas $6,330, Pennsylvania $6,881, Ohio $9,910, with a 10th-to-90th-percentile spread of $4,100 to $12,800.

One caveat on those figures before going further. The card is an image posted to X, and we could not locate a fuller publication behind it. The numbers above are quoted exactly as the card states them. What we cannot verify is anything behind them: the denominator, the cost-center choice, and the hospital set.

The method is what we inferred from Vlada Health’s post. We did not invent it, and nothing below is a claim that we did. Our reply to the thread was the boring caveat: that data is public, and it is also messy and largely unenforced, so the arithmetic is most of the work. This is what happened when we did the arithmetic carefully across our current lineup of data.

The files

File one is the hospital’s machine-readable price file. Most US institutions licensed as hospitals are required by federal law (via CMS, the Centers for Medicare & Medicaid Services) to publish this data. Specified federally operated hospitals, including Indian Health Program hospitals, are exempt or deemed compliant without it. Implant components post in it under HCPCS C1776, “Joint device (implantable)”, which is something of a junk drawer. A big hospital can carry thousands of C1776 lines covering hips, knees, shoulders, trials, screws, and spacers, each one described in whatever abbreviated text its chargemaster happens to use.

File two is the hospital’s Medicare cost report. On Worksheet C Part I, line 72, “Implantable Devices Charged to Patients”, the hospital reports that cost center’s total cost and its total charges. Cost divided by charges is a cost-to-charge ratio. Multiply a posted charge by that ratio and you have an estimate of the cost behind the charge. Not every hospital files line 72. Where it is missing we fall back to line 71, the broader medical-supplies cost center, which happens for 15 of the 319 hospitals below.

A primary total hip needs four parts: a femoral stem, an acetabular cup, a liner that sits inside the cup, and a femoral head. The construct is one representative posted charge for each of those four, added up. The construct cost is that sum times the hospital’s aforementioned cost ratio, line 72 where filed and line 71 otherwise. “C1776” itself is a Healthcare Common Procedure Coding System (HCPCS) Level II code. In medical billing, HCPCS Level I codes are also known as CPT codes, and are owned by the AMA. HCPCS Level II codes are a federal standard, provided in the public domain by CMS, and are commonly used by hospitals and other providers to bill insurance companies. The line descriptions quoted below are not CMS text. They are the hospitals’ own chargemaster wording, published by the hospitals in their own files.

What we found

Across 319 hospitals in ten states, the median implant-construct cost is $6,704.97. The 10th percentile is $4,374.37 and the 90th is $12,357.27, so read that center inside a nearly three-fold band rather than as a single number. The posted charge behind it, before any cost ratio is applied, has a median of $21,332.72.

At the time of this analysis, we have data from ten US states. A state’s numbers publish only when at least five of its hospitals made it into the included set. Seven states clear that bar:

The second column is there because hospitals in one system often post one shared chargemaster, and a state’s hospital count can overstate how many independent files sit behind its cell. More on that below.

Washington, DC and Delaware arrive at three included hospitals each and Wyoming at two. All eight remain in the fleet result. They are not shown as state cells because they have fewer than five included hospitals.

Do those two columns rank together? Not really. Colorado posts constructs 2.3 times Florida’s, but its cost median comes in only about 39% above Florida’s. Colorado’s median cost ratio is 0.2325 against Florida’s 0.3567, and the low ratio absorbs most of the high charge. That divergence between what is charged and what the cost report implies is arguably the more interesting result here.

Colorado is worth one more sentence, because a number that different from the rest in that column deserves an explanation. Its high constructs are concentrated in six HCA HealthONE hospitals, which post constructs from $57,092 to $86,753 on healthy row counts. The printed median of $36,569.58 is the exact construct that four UCHealth hospitals all post, because they share one chargemaster. That is a real posted-charge fact about Colorado chargemasters. It is not a classifier artifact.

One hospital, worked end to end

William P. Clements Jr. University Hospital at UT Southwestern, CCN 450044, in Dallas, carries 8,681 rows under C1776 in the price file we retrieved on June 27, 2026. Every step of its calculation is below, so this can be independently reproduced.

Four representative lines, quoted from its own file, with the posted gross charge each one carries and the number of rows in that component group:

Each posted figure is the median gross charge across that component’s surviving rows. The description beside it is a real line from the file that carries that exact price. The four sum to a posted construct charge of $22,470.10.

The cost report supplies the other half. UT Southwestern’s FY2025 final report, Worksheet C Part I, line 72, cost from column 5 and charges from column 8, gives a ratio of 0.3547. Multiply:

$22,470.10 × 0.3547 = $7,970.14

That is above the fleet median of $6,704.97, on a hospital whose posted construct is also above the fleet median. Both halves of the arithmetic are visible, which is the point of showing one hospital this way. We name this hospital because every posted number above is a fact from its own published file. The derived costs everywhere else in this article are state and fleet aggregates, and this one worked example is the only per-hospital cost figure here.

What the guards do

Two guards run against the data before any median is taken. The penny guard drops rows posting a gross charge of a dollar or less, on the assumption that a hardware charge under a dollar is a placeholder rather than a real price. The self-contradiction guard drops a sub-$100 row when that hospital’s own median for the same component is $1,000 or more. The reasoning: a $54 “femoral head” sitting beside a row of $3,000 heads is probably not a femoral head at all. More likely it is a smaller mislabeled part, like a screw.

Across the fleet the guards dropped 117 penny rows and 177 contradiction rows, and 51 of the 319 included hospitals had at least one row dropped.

Full disclosure. Turning both guards off moves the fleet median by exactly $0.00. No change. It likewise moves six of the seven state medians by nothing. Only Connecticut shifts, from $7,597.01 to $7,352.11, on a seven-hospital cell.

That is not evidence the guards are unnecessary, though it would be easy to read it that way. Per hospital, 28 of the 319 produce a different construct once the guards come off, and the direction is completely one-sided: 28 lower, 0 higher, mean −$500.65. The largest move belongs to Legent Orthopedic Hospital in Texas, whose posted construct drops from $9,350.00 to $7,500.00 when 97 penny rows come back across its four component groups and pull all four medians down. Publishing without the guards produces lower estimates for those 28 hospitals while leaving the headline untouched. Lower is not the same as wrong, and we have no purchase order to check any of them against. The takeaway runs both directions. The fleet median is robust to this choice. Individual hospitals’ numbers are not. Both halves belong on the page.

The contradiction guard has a blind spot worth naming. It compares a suspiciously cheap row against that hospital’s own median for the same component, so it can only catch a bad row sitting beside good ones. A component group that is wrong in its entirety looks internally consistent and passes untouched. One included hospital posts three femoral stems at $22, $123 and $680, and a $123 stem is not a plausible femoral stem, but nothing in the guard can tell that from a hospital whose stems really are cheap.

Sensitivity

Four modeling choices, measured rather than argued:

The published center is $6,704.97 and the axes put it inside a range from $2,036.88 to $22,363.26. Nearly all of that width comes from one choice: which row in a hospital’s list of hundreds of stems represents “a stem”. We use the median because taking the cheapest or the priciest moves the answer by an order of magnitude, and no amount of care elsewhere shrinks that. Line 71, the broader medical-supplies cost center, is a much noisier ratio for this purpose, and forcing every hospital onto it drops 11 that never filed the line. That is why line 72 is the published preference, with line 71 used only where 72 is absent.

Who is not in this

A headline with no stated denominator cannot be checked. Vlada Health’s “236 hospitals in four states” does not say how many were looked at, but ours does. Every hospital in our (as of writing this article) 1,080-hospital lineup lands in exactly one bucket:

The 31 without an extract are our roster’s already-documented acquisition gaps (see our MRF Tracker for details), seven of which are Indian Health Program hospitals lawfully exempt from the posting requirement. The 442 with no C1776 rows are exactly that and nothing more: their extracts contain no C1776 row. We did not measure why, and the obvious guess does not survive contact with the list, which includes hospitals that plainly do joint replacement. Coding practice, device charges packaged into a procedure line, source-file behavior, and service mix would all produce the same silence, and we have not separated them. The 284 that post C1776 without a full construct are the interesting bucket, and the reason this sample is not a random one.

Checking the answer three ways

The whole-operation check, which does not work. Take each hospital’s posted gross charge for CPT 27130, the total hip replacement itself, and multiply by that hospital’s same cost ratio. Across the 44 included hospitals that post a real gross charge for it, the median comes to $3,332.05, or 0.25x the $13,116.76 that Medicare’s CY2026 OPPS rate pays for C-APC 5115. Four times low. The reason is a category mismatch built into the test: line 72 is the implantable-devices cost center’s ratio, fleet median 0.343, and multiplying a whole-operation charge by a device-department ratio has to understate the whole operation. It tells you the line-72 ratios are device-scale. It does not validate whole-operation costing, and we are not presenting it as though it did. Two further limits belong on it. The 44 hospitals are a self-selected 14% slice, because the other 198 included hospitals that post 27130 give it no gross charge at all, only payer rates. And the low tail of the 44 is visibly junk.

CMS’s own device offset, which does work. CMS publishes what share of an APC payment it models as device cost. For APC 5115 in the CY2026 OPPS final rule, Addendum P’s “APC Device Offset Percentage” column reads 44.95% for both 27130 and 27447. Applied to Addendum A’s national unadjusted payment of $13,116.76, that implies $5,895.98 of device cost inside the bundle. Our fleet median of $6,704.97 is 1.14x that figure. Restricting to hospitals with in-band cost ratios gives $6,627.89, or 1.12x. Our 10th percentile sits at 0.74x and our 90th at 2.10x, and 63.6% of included hospitals (203 of 319) land above the CMS number.

That is the anchor that matches: device cost against device cost, built two separate ways. CMS gets there from claims and APC modeling. We get there from chargemasters times cost-report ratios. Landing 14% apart is the strongest corroboration in this run. Both are still allocated-cost estimates rather than invoices, so the agreement is between two accounting models and not between either one and a purchase order.

Florida and Texas against Vlada Health. We committed to reporting this either way:

FL: ours $5,695.85 (n=80)   Vlada $5,401.00   diff $+294.85 (+5.5%)
TX: ours $6,735.56 (n=135)  Vlada $6,330.00   diff $+405.56 (+6.4%)

Both states land above Vlada Health’s published figure by a similar small margin, and the ordering holds, with Texas above Florida in both. Two of the four candidate explanations are ruled out by measurement. Guard choices move the FL and TX medians by $0.00, so no guard explains any of it. Representative choice moves those medians by thousands of dollars, so it cannot produce a consistent few-hundred-dollar offset in the same direction twice. Our file vintage runs June 27 to August 1, 2026 against an August 2026 publication, so vintage is a weak candidate. That leaves two candidates standing: which hospitals made it into each sample, and how the cost-report join lines up on cost center and fiscal year. The evidence here cannot separate them, because without the Vlada Health publication details we cannot see their denominator or their cost-center choice. However, what the comparison does establish is narrower and still worth having: an independently built pipeline over the same two public files reproduces both state medians within 6.4%.

What this cannot tell you; i.e. weaknesses in the analysis

A cost derived this way is Medicare-allowable, overhead-loaded cost. It is an accounting allocation rather than an invoice, and it carries a share of the implant cost center’s indirect expense. It cannot reach the lowest contract prices anybody quotes, and that gap is expected rather than a defect.

The cost ratio, whichever line it comes from, is a department average. Applying one hospital-wide device ratio to four specific SKUs assumes the markup on a femoral stem matches the markup across every implantable device that hospital bought that year, and there is no reason it must.

Requiring all four components selects for hospitals whose files itemize the whole construct. 323 of the 1,080 produced all four components, and 319 of those also had a usable cost-report ratio. 284 fall out at the incomplete-construct step, so this is a sample of well-itemized chargemasters rather than a random sample of hospitals.

Chargemaster descriptions are ambiguous, and the bound on that is measured, not assumed. Our classifier’s precision on the four component classes is 98.3% against a hand-adjudicated fixture, and it declines to name 36.0% of the C1776 rows it sweeps fleet-wide. That 98.3% is an in-sample figure and not a promise about the fleet. An adversarial sample of descriptions drawn from outside the fixture scored lower per string, mostly on rare branded products whose names give no hint that they belong to a revision or a reconstruction. Some of those are now excluded by name. The rest are part of why the classifier declines a row rather than guessing at it. The two failure modes are not equally bad. When the classifier declines to name a row, the worst case is that a hospital drops out of the sample. When it names a row wrongly, a bad number lands in a published price. We chose the first failure over the second on purpose.

The exclusions err toward dropping real parts. “HEMI” on its own is treated as a hemiarthroplasty marker, which is right most of the time and wrong for the occasional cup described by its hemispherical geometry. About 129 descriptions sit in that category awaiting hand adjudication. That costs recall rather than precision, which is the direction we would rather be wrong in.

State counts overstate independence. Of the 319 included hospitals, 82 post the same four component charges as another included hospital’s, because systems share a chargemaster: 19 Texas Health Resources hospitals, 10 INTEGRIS hospitals, 8 Baptist Health South Florida hospitals. That leaves 237 distinct construct signatures behind 319 hospitals, and the state table’s second column carries the per-state count. The cells themselves are medians over hospitals, so a system’s repeated chargemaster votes once per hospital rather than once per file. Give each signature a single vote instead and the posted-construct medians move a long way in the duplicate-heavy states, up 50.6% in Colorado and 68.8% in Texas. The cost medians are far steadier, moving under 1% for Texas and Colorado, about 3% for Florida, and around 11% for Nebraska, Oklahoma and Kansas.

Twenty of the 319 carry cost ratios outside the plausible 0.05 to 0.90 band, one as high as 37.83, which yields a nonsense multi-million-dollar “cost”. Fifteen of those are line-72 ratios and five are line-71 fallbacks, including the 37.83. Medians and percentiles survive them: excluding all 20 moves the fleet median 1.1%, from $6,704.97 to $6,627.89. Dropping every line-71 fallback instead, all 15 of them, leaves 304 hospitals and a median of $6,603.17. The maximum and any mean survive neither cut, which is why neither appears anywhere above and why none of those hospitals is cited individually.

Finally, posted charges are just what the hospital published. They are not quotes, and they are not what you will be billed.

In conclusion, analysis like this can arrive at a reasonably defensible ballpark figure, but there are weaknesses that are presently unavoidable.

What would fix this analysis gap

Put the device identifier on the line. The labels and packages of implantable devices generally carry a unique device identifier under FDA’s UDI rule, and the device records are submitted to FDA’s public GUDID database, which resolves the device portion of that identifier to a brand, a company, a model, and a description, at no cost to anybody. What GUDID does not carry is price, and what the price file does not carry is the identifier. The two public datasets that would answer this question in one join sit next to each other, unjoined, and we are left inferring “femoral stem” from STEM FEM 156MM 18MM PF RDC DIST TPRLK TI PPS 133D HI OFST.

A device identifier on the chargemaster line would make that inference unnecessary. A construct could be priced by the actual part rather than by our reading of abbreviated text, and the classifier precision figure above, and the 36% of rows we decline to name, would both stop being anybody’s problem. Absent that, joining GUDID to chargemaster text means building a text-to-identifier resolver across every vendor’s abbreviation conventions, and that is its own project rather than a footnote to this one.

Vlada Health’s card makes this same ask. Cuban is asking for something adjacent, that the contracts themselves be published, and Slotkin’s contribution was to offer numbers from inside them. Three different directions, pointing the same way. In the meantime, our free file check will tell you whether a given hospital posts its price file and where to find it, and every hospital page on this site links straight to the raw file. Our sources and method are on the about page.