What a Hospital Knows About the Cost of a Hip

In January, Mark Cuban posted that for any hospital claiming to lose money on Medicare and Medicaid patients, he would love to see "the bill of materials and direct overhead attributed to the procedures you lose money on. How much you pay for that hip implant, the anesthesia, disposables, the doctors/nurses etc." He added that "It's crazy that I can't find that information ANYWHERE."

In August he put the general version to KFF. There are a lot of hospitals, he said, "that don't know their costs." They "just know their cash balance."

Most people hold one of two beliefs about this. Either the hospital knows what the hip cost and will not say, or the hospital does not know and ought to.

Here is what the evidence supports. The hospital knows what it paid for the implant, because an implant arrives with an invoice. Standard billing produces no comparable number for the nursing, and the reason is not weak accounting. Under the Medicare cost reporting rules nursing sits inside the daily room rate as a routine service, so every patient on the floor carries the same nursing charge no matter how much nursing that patient used. A hospital can build a patient-specific nursing cost by measuring it, and the ordinary architecture does not hand one over. And when one academic medical center costed its own joint replacements two defensible ways, it got $12,982 and $23,915 for the same hips.

The seen cost of taking a provider's number at face value is one figure you cannot check. The unseen cost is that the same method prices everybody's care. Medicare builds the relative weights behind every inpatient stay in the country out of cost report data derived by this ratio, and CMS's own contractor concluded the method overvalues low-cost diagnosis groups and undervalues high-cost ones. The error does not stop at one hospital's claim about one hip. It is already inside what the country pays for hospital care, in a direction that has been measured for supplies and never for labor.

What he has right

The prevalence is real. HFMA and Strata Decision Technology surveyed 226 C-suite and director-level respondents in the spring of 2019 and found 63 percent of the responding organizations still costing labor with ratio of cost to charges. HFMA's own reading was that progress had largely stalled since HFMA's 2011 research. I found no later published prevalence estimate, which leaves a seven-year-old survey as the most recent figure available.

The group purchasing criticism is also real, and the statute is more permissive than most readers would guess. Under 42 U.S.C. § 1320a-7b(b)(3)(C), a vendor may pay the purchasing agent acting for a group of hospitals, provided a written contract names the amount or the percentage and, where the buyer is a provider of services, the agent discloses to that provider what the agent collected from each vendor.

Cuban told a pharmacy audience in April that hospitals "overpay for everything, literally everything," and that "Your GPOs are not very nice people to you, taking their 3%."

The 3 percent is often described as a cap and it is not one. It appears in the safe harbor at 42 C.F.R. § 1001.952(j) as a drafting threshold. An agreement may state that vendors will pay the purchasing organization 3 percent or less of the purchase price and stop there. Where the fee is not fixed at 3 percent or less, the agreement has to specify the amount, or the maximum amount if the amount is not known. Neither version limits what a vendor may pay.

The annual disclosure appears in both the statutory exception and the regulatory safe harbor, so an arrangement that skips it sits outside both categorical protections and has to be judged under the anti-kickback statute itself, which carries its own elements including knowing and willful conduct. A covered hospital should have that disclosure somewhere in the building. Whether the person making the purchasing decision has ever seen it is a different question.

The Government Accountability Office reviewed the arrangement in 2014 and concluded that almost thirty years after the statutory exception passed, there is little empirical evidence to assess the effect of a funding structure in which the seller pays the buyer's agent. An earlier GAO pilot in 2002 priced pacemakers and safety needles and found the result running both ways. On some pacemaker models, hospitals buying through a group contract paid up to 26 percent less than hospitals that negotiated alone. On other models they paid up to 39 percent more.

Two measurement problems, not two accounting categories

Cuban put two different measurement problems into one sentence, and his own list runs across both.

An implant, a drug and a disposable are purchased inputs. They arrive with a price on a purchase order, and a hospital can trace that price to the case it was used on. The 2019 survey found 68 percent of respondents costing supplies, drugs and implants at patient-specific acquisition cost, with 52 percent of the same respondents also using ratio of cost to charges for supplies, so the two methods run side by side inside the same organizations.

Nursing time and operating room time are different. A nurse's forty minutes on a case is direct labor rather than overhead, and calling it overhead is the first mistake in this argument. Nobody buys those forty minutes on a case-specific invoice. The hospital has to measure how much of the nurse the patient used, or attribute it by some proxy, and the choice between measuring and attributing is where the costing method starts deciding the answer.

Then there is the fact that makes the nursing question harder than the accounting question. Under 42 C.F.R. § 413.53(b), routine services are "the regular room, dietary, and nursing services, minor medical and surgical supplies, and the use of equipment and facilities for which a separate charge is not customarily made." Ancillary services are the ones for which charges are customarily made in addition. Nursing is on the routine side, which means the Medicare cost reporting architecture does not carry a per-case nursing charge at all.

Bonnie Chiang reviewed the methods for estimating nursing cost in 2009 and stated the consequence plainly. Nursing costs are averaged into the daily room rate, so "all patients in a given care unit of the hospital are presumed to consume the same amount of nursing care resources," which "creates a mismatch between resources consumption and billed charges."

So the second half of Cuban's demand runs into something structural. For the largest clinical labor category in the building, the billing system was not built to distinguish one patient from another.

What the ratio actually does

Ratio of cost to charges works like this. Take a department's total costs from the general ledger. Divide by that department's total charges. Multiply any individual charge by the result and you have an estimate of what that item cost.

The obvious objection is that the method is circular, because the hospital sets the charges. There are two versions of that objection and only one of them holds.

If every item in a department carried the same markup, the ratio would recover cost exactly. GAO stated the principle in 2006: "If all services were marked up over costs by an identical percentage, charges would represent the relative costliness of services perfectly." Raise every charge in a department by twenty percent and the department's total charges rise by twenty percent as well, so the ratio falls to five sixths of what it was and every cost estimate lands where it started.

So the version of the objection aimed at the level of charges fails. The version aimed at the structure of charges holds. One ratio applied across a whole department assigns more estimated cost to any item marked up more than its department peers, and less to any item marked up less than its peers. Estimated cost follows position on the department's own markup curve, and the hospital drew that curve.

For supplies, that curve has a documented shape. The Lewin Group interviewed hospitals for MedPAC in 2005. Over half of the respondents used a supply markup table, and all but one of those said the table marks up higher cost items less than lower cost items. CMS's own contractor named the consequence in 2007 and called it charge compression, concluding that low-cost diagnosis groups may be systematically overvalued and high-cost ones undervalued.

A separate problem showed up when Medicare itself used these ratios. Medicare applies a ratio taken from a settled cost report to current charges when it calculates outlier payments, and settling a cost report takes one to three years. In 2003 CMS found that a few hundred hospitals had been raising charges hard against their older ratios to trigger outliers, and changed the rule. That is a defect in the outlier methodology rather than in a hospital's internal cost model, and the two get run together in most tellings.

Two limits on how far the compression result travels, and the second one is the point.

An item-level bias does not give you a case-level total. A case is a stack of charge lines, and the same mechanism overstates the cheap lines while it understates the expensive ones. Whether a particular case nets out high or low depends on its mix, and I did not find anybody who has worked that out.

And the documented markup shape is about supplies and devices. I did not find a study isolating the charge-based allocation of nursing or other clinical labor and establishing which way its error runs. That is the half of Cuban's demand with no invoice behind it, the half where the billing architecture assigns every patient on the floor the same daily amount, and the half where nobody has published a direction of error.

What the hips actually cost

In 2015, Sara Akhavan, Lorrayne Ward and Kevin Bozic published what the rest of this argument has been describing in the abstract. They costed 700 primary joint replacements on 677 patients at one academic medical center over seventeen months, tracing implants and consumables at purchase price and building personnel cost from loaded labor rates against process maps of each step.

For a primary total hip, time-driven activity-based costing produced $12,982. The hospital's existing system produced $23,915. For a primary knee the two figures were $13,661 and $24,796. The categories where the two systems diverged most were operating room services and room and board.

Read what the comparator was, because it is not what a reader expects. The hospital's existing system was itself an activity-based system running in a commercial decision support product, allocating general ledger and payroll expense to activity codes, with indirect cost set at 60 percent of direct cost by rule.

So this is not a sophisticated method beating a crude ratio. It is two activity-based systems at one hospital, run on the same 700 operations, disagreeing by roughly a factor of two on the cost of a hip. The largest gaps sat in the operating room and the room rate, which are the two categories with no purchase invoice behind them.

The authors are careful about what produced their own number. Their process maps rest on "time estimates based on employee and supervisor estimations" rather than direct observation, and they say the method could not fully account for indirect costs such as research and sterile processing.

The method, and the twenty-two years since

Robert Kaplan and Steven Anderson published time-driven activity-based costing in Harvard Business Review in November 2004. Measure how long each step in a process takes, price the capacity that step consumes, and build the cost of the unit from the bottom.

Kaplan and Michael Porter prescribed it for health care in the same magazine in September 2011, writing that "there is an almost complete lack of understanding of how much it costs to deliver patient care." That is Cuban's sentence in different vocabulary, fifteen years earlier.

The literature since then is larger than the critics say and thinner than the advocates need.

A 2017 review of 25 studies found that "All articles conducted their analysis within confines of a department or clinic," that no article assigned the indirect costs of support departments to primary resources, and that the method's ability to inform bundled payment and to coordinate delivery across the care continuum "remains to be demonstrated in the published literature." Those last two are what the 2011 article proposed the method for. Of the 22 hospital studies it found, 17 were in surgical wards.

A 2020 review of inpatient applications found 26 studies, every one surgical, none for medical inpatients, ten of them orthopedic, with "enormous variability" in how the method was applied. It found no studies reporting negative results, and named the implication itself: "there is a potential publication bias toward positive results in this field."

A 2025 review found 32 more, and this one corrects the surgical skew rather than repeating it. Its authors excluded surgery on purpose, writing that "surgical TDABC applications have been extensively reviewed in a recent 2020 review." What they found runs across primary care, secondary care, acute care, tertiary care and long-term management, with seven full economic evaluations among the 32. Anybody claiming the literature has not moved since 2020 has not read the 2025 review.

What has not happened is institutionalization, and one case makes the point better than an absence would.

Cleveland Clinic ran a time-driven costing pilot on mitral valve repair and aortic valve replacement, working with Kaplan and a Harvard colleague, and compared the results against the relative value unit system it had used for twenty-five years. The new method came out about 10 percent lower on both procedures. The team published the result in 2014 and wrote that "we do not believe that these results warrant a wholesale replacement of our current system."

That is an institution with the resources, the partnership and the result, declining to convert. Kaplan himself, writing the same year with clinicians from MD Anderson and Boston Children's, said the approach "has not yet been systematically implemented and evaluated." In 2021, three orthopedic surgeons making the case for it wrote that it "still appears ahead of the innovation curve."

I looked for a health system running the method across the enterprise rather than in a department or a service line, and for a published account of one adopting it and then giving it up. I found neither. Health systems do not publish their cost accounting architectures and vendors sell this software, so a deployment at scale could exist and leave no trace in a journal. What I can say is that nobody has published how common enterprise-level adoption is, how many hospitals have made it the primary cost architecture, or how often it gets abandoned after implementation.

There is also a measurement problem underneath one common way of running the method. Eddy Cardinaels and Eva Labro ran experiments on how people estimate the time a task takes and published in The Accounting Review in 2008. Breaking a job into finer activities made the estimates worse rather than better, and estimates given in minutes carried a strong upward bias. They named the target themselves, writing that their finding "may be problematic for Time-Driven Activity-Based Costing that advocates the use of estimates in minutes." That is a vulnerability in estimate-based implementations. Direct observation and system timestamps are also used, and the 2025 review reports hybrid approaches.

Why it never became the default

A cost system is an investment, and the investment pays when somebody bears the consequence of not knowing.

Eva Labro and Lorien Stice-Lawrence tested a version of that. Across states that enacted hospital price transparency laws, the hazard of a hospital updating its costing and budgeting systems rose, by roughly 14 and 16 percent respectively in the working paper version of their study. What a hospital measures responds to what the rules make it worth measuring.

Here is the objection, and it is the first thing a hospital CFO will say. Prospective payment already creates much of that incentive. For an ordinary Medicare inpatient case, the hospital receives a payment determined largely in advance by diagnosis group, so a dollar of avoidable cost saved is a dollar of margin kept. That system has been in place since the Social Security Amendments of 1983, for cost reporting periods beginning on or after October 1 of that year. It is not one flat amount per case, because Medicare adjusts for area wages, disproportionate share, indirect medical education, high-cost outliers and new technology, and the adjustments matter. The basic incentive has still been running for four decades.

Prospective payment has supplied that incentive for four decades, and the available evidence still does not tell anybody how widely hospitals have installed patient-level costing capable of answering Cuban's question. I do not have an account of why that is good enough to print.

Different costs for different decisions

The reply from inside hospital finance is that most of the cost is joint and fixed, so any per-case number requires an arbitrary allocation, and an arbitrary number is worse than no number. Arthur Thomas made the general form of that argument in 1969 and called allocations incorrigible, meaning nothing in the world can confirm or refute an allocation.

The narrow version of that reply is correct, and it does real work here. J. M. Clark made the point in 1923 under the heading of different costs for different purposes. Mogyorosy and Smith carried it into health care in 2005, writing that "the cost of a particular service can vary substantially according to the purpose of cost data for which it was generated."

Grégoire Mercier and Gérald Naro measured the size of that in 2014, costing 2,130 abdominal procedures two different ways at Montpellier University Hospital. The two methods correlated at 0.73. The range holding 95 percent of the case-by-case disagreements ran from about 2,146 euros in one direction to 2,434 euros in the other, against a hospital average of 2,052 euros per stay. Their verdict was that "the overall agreement between methods was poor."

What that establishes is narrower than the slogan it usually gets turned into. There is no single cost number that answers every question about a hip. Directly consumed resources can be measured, and $4,000 paid for an implant is $4,000 paid for an implant. Shared and capacity costs have to be attributed, and which attribution is right depends on what the number is being used to decide.

That is also where the marginal question and the service line question part company, and they get run together constantly. Eric Noreen and Naomi Soderstrom found that costing models treating hospital overhead as proportional to activity grossly overstate the overhead that is actually incremental, and that predicting a given overhead cost will not change at all, apart from inflation, usually beat predicting it would move with volume. That result says a fully allocated per-case cost is the wrong number for deciding whether one more hip is worth doing inside existing capacity. It does not say the same number is wrong for deciding whether to keep an orthopedic service line, because over a long enough horizon the staff, the space, the equipment and the supervision all become avoidable.

What to ask

If you are reading a provider's claim that a line of business loses money, the bill of materials is the wrong thing to demand. It is the part they can produce, and producing it settles nothing.

Ask three things instead.

What decision is this number supposed to answer, and over what horizon? Whether the next case covers its own cash and whether the service line should exist are different questions with different right answers, and a single per-case figure cannot serve both.

Which resources were traced and which were attributed? For clinical labor and shared capacity in particular, ask what driver was used and why that driver.

What would have to change for the profit or loss to flip sign? A provider who can name it has a decision model. A provider who cannot has an allocation.

Cuban's own hospital-facing work is a purchasing channel and a public library of provider contracts, and the contracts published so far are mostly line-item fee-for-service rates rather than prices for a defined episode. Published prices give a buyer something to compare. They do not give a seller a reason to go measure anything.