Social Queue

Sinclair Says Age Reversal Is Already In Humans: What Is Actually Being Dosed, Why The Endpoint Is Vision Instead Of Aging, And What The FDA Would Need To Approve An Actual Geroprotector

Sinclair Says Age Reversal Is Already In Humans: What Is Actually Being Dosed, Why The Endpoint Is Vision Instead Of Aging, And What The FDA Would Need To Approve An Actual Geroprotector Video Preview 🎧 Podcast episode for paid subscribers only. Also available on Spotify. To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast). Abstract 1. What the podcast claim actually refers to 2. Partial reprogramming in one paragraph, minus the hype 3. Why the eye keeps getting picked first 4. The clock problem, or why methylation age is not an endpoint 5. Aging is not an indication, and the FDA is not going to make it one this decade 6. The stacking strategy every longevity company actually runs 7. Cancer, the question that never survives contact with a podcast 8. Dogs, mice, and the survivorship math 9. The credibility ledger 10. What the money is actually buying 11. Signals worth tracking over the next two years A viral podcast clip claiming that human age reversal technology is "in people right now" is doing a lot of work with very few nouns. The underlying reality is narrower and more interesting than the clip suggests. Key points below: 1. The most plausible referent is a partial epigenetic reprogramming gene therapy delivered by AAV to retinal ganglion cells, using three of the four Yamanaka factors, in an optic neuropathy population with no approved treatment. 2. That trial is legally a vision trial. The primary endpoint will be letters on an eye chart, not years off a clock. No regulator is being asked to bless the word "rejuvenation." 3. Epigenetic clocks are useful research instruments and terrible regulatory endpoints. None meet the bar of a surrogate reasonably likely to predict clinical benefit, and several are noisy enough that a repeat blood draw moves them. 4. The only serious proposal for an aging indication, a composite morbidity endpoint modeled on the metformin trial design, has spent roughly a decade failing to raise about 50 to 75 million dollars. 5. The commercial math for a reprogramming therapy in a rare optic neuropathy looks like ophthalmic gene therapy economics, which is a few thousand patients a year at a six figure price, not a total addressable market of everyone who has ever had a birthday. 6. The cancer question, dedifferentiation being a shared feature of both reprogramming and tumor biology, is the actual gating item, and it is the one that never gets asked on a three hour podcast. What the podcast claim actually refers to A clip is circulating in which a Harvard geneticist tells a comedian that the first potential human age reversal technology is in people right now, and the internet does what the internet does. Almost a million engagements, 863 thousand impressions, a lot of very confident replies from accounts whose bio includes the word "based." Nobody in that thread is asking the boring question, which is what molecule, what route, what dose, what population, what endpoint, and under which IND. Strip the framing and the likely referent is fairly specific. The lab’s most visible human-facing work has been on partial epigenetic reprogramming in the retina, and the company that licensed that work has been developing an optic nerve program through preclinical and regulatory steps toward first-in-human testing. The construct is an adeno-associated virus carrying three transcription factors, OCT4, SOX2 and KLF4, under an inducible promoter so the expression can be switched off. The target is retinal ganglion cells. The indication is nonarteritic anterior ischemic optic neuropathy, which is a mouthful that means the optic nerve suddenly stops working because its blood supply hiccupped, usually overnight, usually in someone over 50, and there is nothing anyone can do about it. So the honest version of the sentence is that a gene therapy which happens to use reprogramming factors is being tested in patients who have gone partially blind. That is a genuinely important trial. It is also not the same claim as "we are reversing human aging," and the gap between those two sentences is where the entire regulatory and investment story lives. Partial reprogramming in one paragraph, minus the hype The four factors won a Nobel because they do something absurd. Put OCT4, SOX2, KLF4 and MYC into an adult cell and hold them there, and the cell forgets what it was and becomes pluripotent. Mouse in 2006, human in 2007, prize in 2012. Along the way to forgetting its identity, the cell also resets its DNA methylation pattern, which means an induced pluripotent stem cell derived from a 70-year-old reads as roughly zero on a methylation clock. That observation is the seed of the whole field. Aging looks, at least in part, like an accumulation of epigenetic noise, and reprogramming appears to sand it off. The obvious problem is that a fully reprogrammed cell in a living body is a teratoma waiting to happen, and a mouse expressing all four factors continuously dies of exactly that within about a week. The fix, demonstrated in a 2016 Cell paper using progeria mice, was to pulse the factors instead of holding them, two days on and five days off, which extended median lifespan by roughly 30 percent in an animal that ages catastrophically fast. Later work dropped MYC entirely, went to three factors, ran long induction schedules in normal mice for seven to ten months without tumors, and reported improvements in skin and kidney readouts. In vitro, transient reprogramming of human fibroblasts for about two weeks knocked something like 30 years off their methylation age while the cells retained fibroblast identity and improved a bit on collagen production and wound healing assays. That last result is the one people quote at dinner parties. It is real. It is also fibroblasts in a dish, where the consequence of a cell losing its mind is that the well gets thrown out. Why the eye keeps getting picked first Every gene therapy program that can plausibly justify starting in the eye starts in the eye, and the reasons are unromantic. The eye is immune privileged, so the systemic response to an AAV capsid is muted. The dose is tiny, microliters instead of milliliters, which means manufacturing burden per patient is low and the systemic exposure is negligible. The other eye is a built-in control. The tissue can be imaged at micron resolution in a five-minute clinic visit, so retinal nerve fiber layer thickness and ganglion cell layer thickness become continuous readouts rather than biopsies. Subretinal and intravitreal delivery of AAV is a well-trodden path with approved products behind it. And if something goes badly wrong, the failure mode is contained inside a structure roughly the size of a ping pong ball. The indication choice is equally strategic. Nonarteritic anterior ischemic optic neuropathy affects somewhere in the range of two to ten people per hundred thousand over 50 in the United States, call it several thousand new cases a year, which comfortably clears orphan designation. There is no approved therapy. The last serious pivotal attempt at a neuroprotective agent missed. Patients present with sudden painless vision loss and are told to manage their vascular risk factors and hope. That combination, no standard of care and a measurable deficit, is the friendliest regulatory environment in ophthalmology. There is also a grimly funny tailwind. The same condition picked up a safety-signal association with GLP-1 receptor agonists in a 2024 retrospective from a major eye hospital, and European regulators added nonarteritic anterior ischemic optic neuropathy to semaglutide labeling as a very rare adverse reaction in 2025. So the first indication for the first human reprogramming therapy is a rare optic neuropathy whose profile got raised by the best-selling drug class of the decade. The longevity field has been waiting for a demand catalyst and it arrived via Ozempic. The clock problem, or why methylation age is not an endpoint Here is where the technical audience should get itchy. The rhetorical engine of longevity biotech is the epigenetic clock, and clocks are much weaker instruments than the marketing implies. The first-generation clock, published in 2013, used a few hundred CpG sites to predict chronological age with a correlation near 0.96. That sounds spectacular until you notice what it means. A model trained to predict chronological age is, at best, a very good chronological age predictor, and its residual is where any biology lives. Second-generation clocks fixed some of this by training on mortality and clinical chemistry instead, which is why the mortality-trained versions outperform on hard outcomes. A third framing abandoned static age entirely and estimated pace of aging from longitudinal cohort data, producing a rate rather than a number. None of these were built to be moved by a drug and then interpreted. Technical reliability is a live problem. Several widely used clocks show poor test-retest agreement on the same sample run twice, which is why principal component versions were developed specifically to denoise them. Cell composition confounds blood-based estimates, so a shift in neutrophil-to-lymphocyte ratio from an unrelated infection can walk the number around. And the direction of movement is not self-evidently good. Acute injury, inflammation, and even some interventions that shorten life in animals can push clock outputs in the direction everyone calls "younger." For regulatory purposes the bar is explicit. A surrogate endpoint supporting accelerated approval has to be reasonably likely to predict clinical benefit, and getting there requires a chain of evidence linking the marker to outcomes across multiple interventions. LDL cholesterol took decades and several drug classes. Bone mineral density took a landmark fracture trial and is still argued about. Viral load and HbA1c earned their status through repeated demonstration that moving the marker moved the disease. Methylation age has none of that. There is no trial in which an intervention lowered a clock and subsequently reduced mortality in the same population. Until that exists, a clock is a nice figure in the appendix. Aging is not an indication, and the FDA is not going to make it one this decade Drugs get approved for indications. Indications are diseases. Aging is not a disease in any coding system that matters, which is not a philosophical claim but an administrative one. The international classification briefly flirted with a code for old age and then pulled it after geriatricians objected that it would pathologize being 80. There is an extension code for age-related conditions, which is a different thing entirely and does not create a payable indication. The only serious workaround anyone has designed is a composite morbidity endpoint. The metformin trial concept, in development since roughly 2015, proposes enrolling around three thousand older adults and measuring time to first occurrence of any of a bundle of outcomes, myocardial infarction, stroke, cancer, dementia, or death, on the theory that a drug slowing aging should delay all of them simultaneously rather than any one specifically. Regulators have been reported as receptive to the design as a proof of concept. The trial has never fully funded. The ask is somewhere in the 50 to 75 million dollar range over six years, which is a rounding error next to the multi-billion-dollar financing that walked into a single reprogramming company in 2022, and it remains unfunded because nobody owns metformin and therefore nobody captures the upside of proving it works. That single fact explains more about the field than any mechanism slide. The capital is not allocated to answering the question. It is allocated to owning the asset. Function-based endpoints are the other candidate, and they are underrated. Gait speed, grip strength, chair stand time, and composite physical performance batteries all predict mortality well, and the agency has engaged with physical function endpoints in the sarcopenia setting for years without ever approving anything, partly because the field cannot agree on a case definition. Immune function endpoints, vaccine response in particular, have been used in trials of mTOR inhibitors in older adults with reasonable results and no approval. There is a path here. It is slow and it does not make for a good podcast segment. The stacking strategy every longevity company actually runs Given all of the above, the operating playbook is completely rational and almost never stated plainly. You do not seek an aging indication. You pick a specific disease where your mechanism has an outsized effect and a measurable endpoint, run a normal trial, get a normal approval, and then repeat in adjacent indications until the label sprawl adds up to something that looks like a geroprotector without ever using the word. Ophthalmology first, because of everything described above. Then possibly other optic neuropathies including glaucoma, where the patient population jumps by orders of magnitude and the endpoint infrastructure already exists. Then the liver, which is the tissue several reprogramming companies have prioritized for systemic delivery because hepatocytes are accessible to lipid nanoparticles and liver function has clean chemistry readouts. Then immune cells, where the rejuvenation thesis is most mechanistically defensible and where a functional endpoint like vaccine response or infection incidence in the elderly is achievable. Skin sits off to the side as the aesthetics wedge, which is where transient mRNA-based reprogramming has been pointed, because dermatology tolerates trial designs and price points that systemic medicine does not. The financial implication for anyone underwriting this space is uncomplicated: rare, procedure-based, and biologically constrained programs can support premium pricing, but they do not by themselves create a geroscience franchise. The market is still buying a disease asset first and a longevity story second. Cancer, the question that never survives contact with a podcast The deepest unresolved issue with reprogramming is not whether the clocks move. It is whether pushing cell identity backward without releasing control creates more cancer risk than aging risk it removes. The biology is not subtle. Dedifferentiation, proliferative plasticity, and epigenetic loosening are features shared by both rejuvenation and malignancy. The field’s most reassuring data come from transient or partial protocols that keep exposure short enough to avoid full identity loss, but that is not the same as proving safety in a human body that is supposed to live for decades after treatment. That is why the eye matters again. A local treatment in a small, measurable compartment is far easier to justify than systemic delivery. It is also why any serious discussion of human rejuvenation has to be about dose, duration, tissue, and reversibility, not slogans. The first programs will be judged on whether they can preserve or restore function without triggering uncontrolled growth. That is a standard therapeutic question, not a philosophical one. Dogs, mice, and the survivorship math The animal data are persuasive in the narrow sense that they justify more animal work and cautious human trials. They are not persuasive in the broader sense that would be needed to claim an actual geroprotector. Progeroid mice are useful because they fail fast, but they also overstate effect sizes that do not necessarily translate to normal aging. Normal mice are better, but they are still mice, with short lifespans and artificial induction schedules. Dogs are attractive precisely because they age on a more human time scale, but the trials are expensive, slow, and underpowered for mortality endpoints. That survivorship math matters. A therapy that looks dramatic in a mouse can be commercially and biologically modest in humans if it only helps a small number of rare patients or if the treatment effect is too subtle to register against the background rate of aging. The field keeps oscillating between universal claims and orphan-disease execution because those are the two places where the data and the capital are easiest to assemble. The credibility ledger If you want to know whether the field is maturing, track what is being claimed versus what is being measured. The claims are getting bigger. The measurements are still ordinary clinical ones: visual acuity, retinal thickness, functional scores, chemistry panels, adverse events, and eventually perhaps biomarkers that can be correlated with outcomes. That is healthy. What is not healthy is the habit of collapsing a local, mechanism-linked intervention into a universal aging cure. The science may eventually support broader use. The current evidence does not. What the money is actually buying The money is buying platform optionality, not proof of immortality. It funds vector engineering, tissue targeting, biodistribution, manufacturing, toxicology, and the long, expensive work of turning a mechanism into an approvable product. It also funds narrative control, which in this corner of biotech is almost as important as chemistry. If the program works in the eye, it buys a second act in adjacent ocular disease. If it works in skin, it buys a consumer-adjacent entry point. If it works in liver or immune cells, it buys a much larger therapeutic canvas. None of those outcomes requires the FDA to approve aging as a disease, and none of them guarantees that the company’s most ambitious claims will survive contact with the clinic. Signals worth tracking over the next two years The most important signal is whether the first human trial reports a clean safety profile with a meaningful visual endpoint. The second is whether the company expands into another tissue with a similarly ordinary clinical endpoint rather than leaning harder on clocks and press language. The third is whether any sponsor manages to fund a serious aging-endpoint trial without wrapping it inside a disease indication. If that happens, the field will have moved from aspiration to evidence.

CMS's RAPID Coverage Pathway for Breakthrough Devices: What the August 2026 Procedural Notice Actually Does, Why TCET Got Paused, and Why the First Same-Day NCD Probably Lands After 2030

CMS's RAPID Coverage Pathway for Breakthrough Devices: What the August 2026 Procedural Notice Actually Does, Why TCET Got Paused, and Why the First Same-Day NCD Probably Lands After 2030 Video Preview 🎧 Part I Podcast free on Spotify. 🎧 Part II Podcast episode for paid subscribers only. Also available on Spotify. To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast). Abstract 1. Three brands, one unsolved problem 2. What the notice actually says 3. The eligibility gate is narrower than the press release 4. The IVD carve-out, or how MolDX kept its kingdom 5. Coverage is not payment, and payment is where the year goes 6. The same-day NCD is hostage to FDA's publishing calendar 7. CED is where the Class III devices go to sit down 8. Trial economics: what it costs to design for two agencies 9. The free option, and the adverse selection it invites 10. Medicare Advantage, commercial followers, and the real prize 11. Timeline math: when the first RAPID NCD actually shows up 12. What belongs in a comment letter before the window closes 13. Where this leaves builders and capital • CMS published CMS-3487-NC on August 11, 2026, a notice with comment period establishing the Regulatory Alignment for Predictable and Immediate Device coverage pathway. Comments close 60 days out, roughly October 10. • The core promise: a proposed NCD posted the same day as FDA market authorization, 30-day comment period, final NCD around 60 days post-authorization for Class II and 90 days for Class III. Compare to the 9 to 12 months the NCD process normally eats. • Eligibility is tight: presumptive Class II Breakthrough devices in FDA's TAP heading for a De Novo, or Class III Breakthrough devices heading for a PMA, and only at the IDE pre-submission stage. In vitro diagnostics are excluded outright. • TCET is paused for new candidates as of publication. Parallel Review survives. RAPID NCDs jump the NCD wait list. • The structural catch: pre-submission-only entry means the first device to actually complete the pathway is a late-decade event, unless CMS adopts the temporary on-ramp it explicitly asks about. • The other structural catch: an NCD is a coverage decision, not a payment rate. Coding and rate-setting cycles still run on their own clocks. 1. Three brands, one unsolved problem The gap between FDA market authorization and Medicare coverage has been the medtech industry's favorite complaint for about fifteen years, and it has now generated three separate federal attempts to fix it, each with a different acronym and each announced with the confidence of someone who has not read the previous two. MCIT arrived in January 2021, would have handed every Breakthrough-designated device four years of automatic national coverage, and was repealed that November on the grounds that automatic coverage for devices never tested in older patients with four comorbidities is a strange way to run an insurance program. TCET landed in August 2024, promised a final NCD within six months of market authorization, and then capped itself at about five candidates a year, which for an FDA program that has issued north of a thousand Breakthrough designations since 2015 is less a pathway than a lottery. RAPID showed up on April 23, 2026 as a joint CMS and FDA announcement and became a real document on August 11, 2026. So the problem is genuinely hard, and the industry's framing of it has always been slightly self-serving. The framing goes: FDA cleared it, therefore Medicare should pay for it. CMS has been consistent for two decades that these are different questions under different statutes. FDA asks whether a device is safe and effective for its labeled indication. CMS asks whether the thing is reasonable and necessary for the diagnosis or treatment of illness or injury in a population that skews 65-plus, where roughly half of patients carry two or more chronic diseases, and where the pivotal trial that got the device authorized may have excluded exactly those people because comorbidity muddies an effect size. That is not bureaucratic obstinance. That is two agencies with two statutes and one shared inability to make the other one's problem go away. What makes RAPID interesting is that it does not try to resolve the tension by fiat, the way MCIT did. It resolves it by moving the conversation about four years earlier, to the moment when a manufacturer is still sketching an IDE protocol and has not yet spent the money. That is the right instinct. Whether the execution survives contact with coding cycles, benefit categories, and the calendar is the rest of this essay. 2. What the notice actually says Strip the preamble and the pathway is three stages. Stage one is IDE pre-submission. A manufacturer that already has Breakthrough designation, and TAP enrollment where applicable, emails FDA's TAP program before filing an IDE pre-sub and says it wants in. FDA screens for initial eligibility, loops in CMS, and CMS runs a preliminary benefit category assessment. Then there is a kickoff meeting with all three parties where the manufacturer walks its clinical study synopsis and CMS and FDA give live reactions. The manufacturer then requests formal written feedback, which comes back as combined FDA and CMS comments on whether the proposed endpoints will actually support an NCD, whether the CMS IDE criteria are satisfied, and where the evidence gaps sit. Stage two is the formal IDE submission. FDA reviews on its normal clock and issues a decision letter that includes study design considerations. FDA shares that letter, considerations included, with CMS. CMS then talks directly to the manufacturer about which considerations must be addressed to stay in the pathway, and once those are resolved and FDA has blessed any revised protocol, the manufacturer submits the protocol to CMS through the existing centralized IDE review process that has been running since 2015 under 42 CFR 405.212. If everything checks out, CMS issues an approval letter that includes a stated intent to issue a proposed NCD concurrently with market authorization. Any subsequent protocol change needs both agencies to sign off or the device falls out. Stage three is the transition to coverage. FDA shares the IDE final report with CMS. When the marketing submission is accepted for review, CMS gets the clinical study report. The manufacturer sends a formal NCD request cover letter. On the day FDA grants market authorization, CMS posts a tracking sheet and a proposed NCD, opens a 30-day comment period, and targets a final NCD at roughly 60 days post-authorization for Class II and 90 for Class III. Coverage is limited to the FDA-authorized indications. If the evidence has gaps, the NCD comes out under coverage with evidence development instead of straight 1862(a)(1)(A) coverage. Worth noting that none of this requires new statutory authority, which is why it is a procedural notice and not a rule. The 60 and 90 day targets fit inside the existing NCD statutory clock, which allows 60 days after the close of the comment period to issue a final decision. RAPID is not bending the law. It is doing the analytical work in advance so the clock starts from a running start instead of a cold one. The flip side of not needing rulemaking is that the whole thing can be undone by a future notice on a Tuesday afternoon, with no rulemaking record, no appeal rights, and nothing for a manufacturer to sue over if CMS decides in year three that the pathway is inconvenient. 3. The eligibility gate is narrower than the press release Read the criteria closely and the addressable population shrinks fast. Class II devices qualify only if they hold Breakthrough designation, are enrolled in FDA's TAP program, and plan to submit a De Novo request. Class III devices qualify if they hold Breakthrough designation and plan to submit a PMA, TAP participation optional. There is a footnote that lets in 510(k)-cleared devices where the primary predicate was authorized via De Novo no earlier than 18 months before acceptance, which is a narrow window designed to catch second-movers in a brand new device category and not much else. Everything else is out. Then comes the timing gate, which is the one that matters most. The device must be at the IDE pre-submission stage. Not market-authorized. Not mid-study. Pre-submission. If the IDE study is already enrolling, the door is closed, and CMS says plainly that such devices belong in a conventional NCD request, an LCD, or claim-by-claim adjudication. CMS knows this is the binding constraint, because it explicitly solicits comment on whether to build a temporary process letting devices already under an active IDE become eligible, subject to an assessment of whether the running study evaluates the right outcomes or could be amended to do so. That request for comment is the single most consequential paragraph in the document for anyone with a device in the clinic today. The quieter constraint is the phrase separately payable. The device has to be something that can be billed to Medicare on its own if approved. That knocks out an enormous share of implantables and procedural devices whose cost is packaged into a DRG or an APC and never appears as its own line. A transcatheter valve, a neurostimulator lead, a surgical mesh: these get paid through the procedure, not as separately payable items, absent a pass-through or new-technology add-on designation. So the pathway naturally biases toward durable medical equipment, devices with their own HCPCS identity, and technologies that will plausibly qualify for pass-through status. That is a real category, but it is not the whole Breakthrough pipeline, and the notice does not spell out how the separately payable test gets applied at the pre-submission stage when nobody knows the coding outcome yet. The benefit category screen is deliberately weak, and the weakness is honest. CMS commits only to checking whether information exists that immediately makes clear the device will not fall under a benefit category. Double negative on purpose. It is a screen for obvious disqualification, not an affirmative determination, and the notice says outright that acceptance into RAPID should not be read as a final benefit category determination. Anyone building software-as-a-medical-device, a digital therapeutic, an AI-enabled monitoring product, or anything else that lives in the statutory dead zone between the Part B benefit categories should treat that sentence as a flashing light. Getting a fast NCD for something with no benefit category is like getting an expedited visa to a country that does not exist. 4. The IVD carve-out, or how MolDX kept its kingdom CMS excluded in vitro diagnostics from RAPID entirely. The stated reasoning is that FDA's statutory definition of device sweeps in IVDs including diagnostic laboratory tests, that coverage policy for these products is a specialized area, and that CMS has historically delegated the review to particular Medicare Administrative Contractors who should continue to handle it. Translated: the molecular diagnostics coverage program run out of the contractor world is not going to be disturbed, and CMS is not interested in opening a national coverage front on lab tests. This is a bigger deal than the two paragraphs it gets. A meaningful share of Breakthrough designations have gone to diagnostics, including a lot of the AI-driven and multi-analyte products that generate the most investor enthusiasm. Every one of them is now formally outside the fast lane and back in the world of technical assessment submissions, contractor-level policy, and pricing that runs through the clinical laboratory fee schedule with its own gapfill and crosswalk drama. The notice leaves a door cracked, saying that in rare cases where a manufacturer and CMS agree an NCD is appropriate for an IVD, a conventional NCD request can still be filed. Nobody should build a plan around that sentence. There is a defensible policy argument here. National coverage determinations are blunt instruments for diagnostics, where clinical utility varies enormously by indication, by line of therapy, and by what the clinician does with the result. A contractor-level program can move faster and iterate more than the national process, and it does. There is also a less flattering reading, which is that a national fast track for diagnostics would have created immediate pressure on the pricing side, and pricing for lab tests is a fight CMS has been losing in a slow, grinding way for years. Either way, the diagnostics community should read the exclusion as a signal about where the agency intends to spend its scarce coverage staff, and plan accordingly. 5. Coverage is not payment, and payment is where the year goes Here is the part that will disappoint anyone reading the fact sheet without a reimbursement person in the room. An NCD says Medicare covers the thing. It does not say what Medicare pays, or under what code, or in which setting, or starting when. Those are separate processes on separate calendars, and none of them accelerate just because a coverage decision landed early. A device that needs a new HCPCS Level II code enters an application cycle with published deadlines and preliminary and final decision meetings. A device seeking a new technology add-on payment in the inpatient setting applies against the IPPS rulemaking calendar, which means an application filed in the fall gets proposed in the spring rule and effective the following October first, assuming it clears the substantial clinical improvement test, which has its own evidentiary standard that overlaps with but is not identical to the coverage standard. A device seeking transitional pass-through status in the outpatient setting applies on a quarterly cycle and, if granted, gets a limited window of separate payment before the cost is folded back into the APC. Physician work associated with a new procedure needs a CPT code, which runs through an editorial panel with its own meeting calendar, and then valuation, which runs through a different committee and lands in the physician fee schedule rule. Stack it up and a device can hold a final NCD in month three post-authorization and still be waiting on a workable payment rate in month fifteen. The notice does not address this, which is a real omission given that TCET at least gestured at coordinating benefit category, coding, and payment reviews as part of its design. RAPID says nothing comparable. If the agency wants the pathway to change commercial outcomes rather than just press releases, the coding and payment side needs a companion process, and the comment period is the moment to say so in writing with specific deadline alignments proposed rather than a general plea for coordination. 6. The same-day NCD is hostage to FDA's publishing calendar The centerpiece commitment, proposed NCD posted the same day as market authorization, has a dependency the notice states plainly and then moves past. Timing is contingent on the relevant FDA Decision Summary for a De Novo or the Summary of Safety and Effectiveness Data for a PMA being publicly available on the day of authorization. CMS will link to it from the tracking sheet. This matters because CMS is required to build an NCD on publicly available information, and the confidential IDE final report that FDA shares with CMS cannot be the public evidentiary basis. The Decision Summary or SSED has to carry the weight. Anyone who has waited for an SSED knows these documents have not historically appeared the same day as the approval letter. They appear when they appear, sometimes weeks later, occasionally longer, and their content reflects an internal negotiation about what gets disclosed. RAPID therefore requires a change in FDA publication practice, not just a change in CMS analytical sequencing, and the notice does not describe how that change gets operationalized or what happens when the summary slips. A one-week slip turns a same-day proposed NCD into a one-week-late proposed NCD, which is fine. A two-month slip while a redaction dispute gets resolved turns the whole timeline into a normal NCD with better branding. There is a second-order effect worth watching. The SSED becomes a load-bearing coverage document rather than a transparency artifact. Manufacturers will start negotiating its contents with an eye toward what CMS needs to see, particularly around subgroup results in the Medicare-eligible population. Regulatory affairs teams that have historically treated the SSED as an afterthought at the end of an exhausting review will need to treat it as the first page of the coverage file. That is a genuine workflow change and a genuine consulting opportunity for whoever gets there first. 7. CED is where the Class III devices go to sit down The notice is refreshingly honest that a chunk of RAPID NCDs will land under coverage with evidence development rather than full coverage. The logic it offers is risk-stratified: lower-risk devices are more likely to have generated enough evidence by market authorization to satisfy the reasonable and necessary standard, while higher-risk devices are more likely to have residual gaps requiring more evidence under the separate CED authority. Since Class III PMA devices are by definition the higher-risk cohort, the implication is that the 90-day Class III timeline frequently delivers a CED decision rather than open coverage. CED is not nothing. It is coverage. But operationally it is a different animal. Coverage of related services begins once a CED study is approved, which means the final NCD is not the last gate. There is a protocol, there is CMS review of that protocol, and there is review of the CED framework by the health services research agency that has statutory responsibility for defining the research standards. Approved studies then appear on a CMS web page, and providers who want to furnish the device have to be participating in one. Anyone who lived through the transcatheter aortic valve registry buildout knows what that means for diffusion: coverage exists, but it exists at participating centers, with data submission obligations, with credentialing and volume expectations layered on top by the specialty societies, and with a long tail before the technology reaches a community hospital. The one meaningful improvement RAPID offers here is the commitment to engage before market authorization so that CED requirements align with any FDA-required post-approval study. Running one study that satisfies two agencies instead of two studies that satisfy one each is real money and real time. That is a genuine efficiency, and if it holds, it is arguably a larger contribution than the same-day NCD headline. The notice also emphasizes that CED is meant to be time-limited, that a cycle completes when CMS reconsiders and drops the study participation requirement, and that any member of the public can request a reopening. Whether that discipline materializes is an empirical question with a discouraging track record. Several CED policies have outlived the technologies they were built to study. 8. Trial economics: what it costs to design for two agencies The price of admission to RAPID is designing an IDE study that enrolls Medicare beneficiaries and measures clinical outcomes, defined the old-fashioned way as direct measures of how a patient feels, functions, or survives. Surrogate endpoints, biomarker shifts, and imaging changes are not what CMS means. This is a substantive constraint with a substantive cost. Enrolling older patients with multiple chronic conditions raises screening burden, raises dropout, raises adverse event rates that have nothing to do with the device, and can dilute a treatment effect that looked clean in a healthier cohort. Powering for a credible read in the 65-plus stratum, rather than just permitting those patients to enroll, means a larger trial, which means more sites, longer enrollment, and more money. Functional and survival endpoints usually mean longer follow-up than the mechanistic endpoints a sponsor might otherwise pick. For an IDE pivotal that already runs somewhere in the low tens of millions for a Class III implantable, the Medicare-facing design changes plausibly add twenty to forty percent to cost and six to eighteen months to timeline, before anyone accounts for the risk that the harder endpoint simply fails. That last risk is the one boards should be modeling explicitly. There is a real scenario in which the study design that maximizes coverage probability reduces authorization probability, because a functional endpoint in a sicker population is harder to hit than the endpoint the sponsor would have chosen alone. The notice's mechanism for handling this, FDA determines which clinical outcomes are appropriate and CMS confirms whether evidence on those outcomes in the Medicare population would demonstrate improvement, is a reasonable attempt to keep the two agencies from pulling in opposite directions. It is still two agencies with veto power over a protocol, and the manufacturer is the one financing the compromise. The calculus is straightforward for a company whose Medicare mix is 70 percent, which describes most of structural heart, electrophysiology, spine, wound care, and a lot of DME. It is much less obvious for a company whose Medicare mix is 25 percent and whose commercial payers have historically moved on their own timeline. For that second group, RAPID asks for a large investment in a trial redesign to accelerate a coverage decision for a minority of revenue. 9. The free option, and the adverse selection it invites Buried in the general principles is a provision that changes the strategic character of the whole pathway. A manufacturer may withdraw from RAPID at any point up until CMS issues a proposed NCD, for essentially any reason, with incomplete or insufficient data given as an example, and with the alternative of pursuing local coverage explicitly named as a legitimate motive. Think about what that creates. A sponsor enters the pathway, gets four years of privileged access to CMS thinking, gets written feedback on evidence gaps, gets its protocol shaped by both agencies, runs the study, and then looks at the unblinded results. If they are strong, it files the NCD request letter and takes the same-day proposed NCD. If they are ambiguous, it withdraws and quietly pursues an LCD or claim-by-claim adjudication, where the evidence bar is applied by a contractor rather than by the national coverage group, where there is no 30-day national comment period for competitors and specialty societies to file objections, and where there is no permanent national document memorializing the fact that CMS looked at the data and was unimpressed. That asymmetry is rational for the sponsor and awkward for CMS. A negative national coverage determination is a company-ending event in a way that a quiet retreat to contractor-level coverage is not. So the population of devices that actually reach the RAPID NCD stage will be selected for strength, which is fine for beneficiaries and fine for the agency's success metrics, but it means the pathway's eventual batting average will overstate its effect. The devices that would have been genuinely uncertain calls will not be in the sample. Anyone evaluating RAPID's performance in 2032 should ask how many entrants withdrew and where they went, not just how many NCDs got issued in 60 days. There is a countervailing lever. CMS and FDA reserve the right to eject a manufacturer that fails to provide requested information or is found falsifying or omitting material facts. That handles bad faith. It does not handle rational withdrawal, and it is not meant to. 10. Medicare Advantage, commercial followers, and the real prize Roughly 51 percent of the beneficiary population sits in Original Medicare as of 2025, about 34 million people out of nearly 70 million. The other half is in Medicare Advantage, and MA plans are obligated to furnish the Part A and Part B benefits, which means an NCD binds them too. A same-day proposed NCD followed by a 60-day final therefore reaches the entire Medicare population, not just the fee-for-service half, which is a substantially better outcome than a favorable LCD covering one contractor's jurisdiction. The nuance is utilization management. Where national coverage criteria are fully established, MA plans have limited room to layer their own internal criteria on top, and recent policy has tightened that further. Where the NCD comes out under CED, the picture gets murkier, because participation in an approved study becomes a condition of coverage and MA network configurations may or may not include participating centers. That is an unresolved operational question the notice does not touch, and it is a legitimate comment topic for anyone with an MA-heavy account base. The larger prize sits outside Medicare entirely. Commercial medical policy departments do not follow NCDs automatically, but they read them, and a favorable national determination with a public evidence summary is the single most useful artifact a device company can put in front of a commercial medical director. The historical pattern is a lag of somewhere between six and twenty-four months from a positive NCD to broad commercial policy alignment, faster where specialty society guidelines move in parallel. Pulling the NCD forward by nine months pulls the commercial cascade forward by roughly the same amount, which compounds. For a venture-backed device company burning eight figures a year while waiting for the reimbursement flywheel to catch, that is the difference between one more round and an acqui-hire. 11. Timeline math: when the first RAPID NCD actually shows up Do the arithmetic on the pre-submission-only entry rule. A company that decides today to enter RAPID needs Breakthrough designation in hand, which takes months if it does not have one, and TAP enrollment if it is on the Class II track. Then comes the interest email, eligibility screening, the preliminary benefit category assessment, the kickoff meeting, the written feedback request, and the protocol revisions that follow. Call that six to twelve months of pre-submission work, which is not pessimistic for a process involving two agencies and a document that both must effectively agree on. Then the formal IDE goes to FDA, which reviews on its normal clock and issues study design considerations. Those get addressed, FDA blesses a revised protocol, and only then does CMS run its own IDE review. Call that another six to nine months in the good case. Now the pivotal starts. A Class III implantable trial with functional or survival endpoints and a Medicare-representative population is two to four years from first enrollment to database lock, and that is before follow-up requirements that often extend to twelve or twenty-four months post-procedure. Then the PMA gets filed and reviewed, historically the better part of a year. Add it up and the earliest plausible completion for a device entering the pathway from a standing start in late 2026 is somewhere in 2031 or 2032. Even a company that already has designation, already has TAP, and already has a protocol nearly ready cannot realistically beat 2030. Which means the administration announcing this pathway will be out of office before it produces its first same-day NCD, and the fact sheet's promise of accelerated access describes something that happens after the next two presidential elections. This is precisely why the comment solicitation on a temporary on-ramp for devices already under an active IDE is the whole ballgame. If CMS builds that on-ramp, devices currently mid-study could amend protocols or demonstrate that existing endpoints already satisfy the requirement, and the pathway could produce output in 2028 or 2029 instead. If CMS does not, RAPID is a policy commitment to a successor administration that has no obligation to honor it, and the odds of it surviving intact through a transition are not good given that its two predecessors did not. 12. What belongs in a comment letter before the window closes Comments are due 60 days after the August 11 publication, filed under file code CMS-3487-NC. Four topics deserve the ink. The on-ramp question deserves the most, and the answer should be yes with specifics rather than yes in principle. A useful letter proposes a defined window, describes the criteria under which an in-flight IDE could be deemed adequate without amendment, and addresses what happens when a partial amendment is needed mid-enrollment, including whether previously enrolled subjects count. Vague enthusiasm will not move a drafting team; a proposed operational test might. The coding and payment gap deserves the second most. A letter that documents actual elapsed time from NCD to workable payment rate for two or three recent technologies, and proposes calendar alignment between RAPID acceptance and the relevant coding application deadlines, is worth more than a hundred letters asking CMS to consider coordination. The point is that a same-day coverage decision paired with a fifteen-month wait for a rate is not access. The separately payable criterion deserves clarification, because applying a payment-status test at the pre-submission stage, years before any coding determination, is either meaningless or arbitrary depending on how it gets implemented. Manufacturers should ask CMS to state what evidence establishes separate payability prospectively and whether a good-faith intent to seek pass-through or add-on status suffices. The IVD exclusion deserves a challenge from the diagnostics community, not to overturn it wholesale but to define the rare case where an NCD is appropriate. As written, that exception is undefined and therefore unusable. A workable version would identify categories, for instance a diagnostic whose result gates access to a therapy already covered under an NCD, where national consistency has obvious value. 13. Where this leaves builders and capital For operators, the practical guidance is unglamorous. If a device is Class III, Breakthrough-designated, Medicare-heavy, and has not yet filed an IDE pre-submission, RAPID is worth serious modeling and probably worth entering, because the pre-submission feedback alone is valuable even if the pathway is later abandoned. If a device is mid-study, the correct action this quarter is a comment letter arguing for the on-ramp, followed by a conversation with CMS about conventional options. If the product is an IVD, none of this applies and the contractor route remains the route. If the product is software without a clean benefit category, the coverage timeline was never the binding constraint, and the honest answer is that the statutory gap needs Congress, not a procedural notice. For investors, the underwriting change is narrower than the headline suggests but not trivial. The classic device valley of death, where a company survives clearance and then dies waiting for coverage, gets shallower for a specific slice of the market: Class II and Class III Breakthrough devices with separate billability and heavy Medicare exposure. For that slice, coverage risk moves earlier in the diligence stack and becomes partly assessable at Series A rather than a black box discovered at Series C. Diligence should now ask whether the pivotal is designed to satisfy CMS as well as FDA, whether Medicare beneficiaries are enrolled in numbers sufficient to read, and whether management has had the pre-submission conversation. A no to all three in a Medicare-dependent category is a finding, not a footnote. The larger caution is that this is the third federal attempt at the same problem in six years, delivered through the least durable instrument available, dependent on an FDA publishing practice that has not yet changed, silent on the payment machinery that consumes most of the elapsed time, and structurally incapable of producing an output before the end of the decade unless CMS accepts the on-ramp it just asked about. That is a lot of conditional. It is also, on balance, better designed than either predecessor, because it puts the coverage conversation at the moment when a protocol can still be changed cheaply rather than at the moment when the data are already collected and everyone is arguing about what they mean. Aligning expectations before the money is spent is the correct instinct. Whether the federal government can hold that instinct steady for six years across an election is the actual open question, and no procedural notice has ever answered it.

Organ Procurement Under Federal Fire: How Cost-Plus Reimbursement, a 2020 Metric Rewrite, and the DCD Boom Turned Death Determination and Organ Allocation Into Healthcare's Next Compliance Fight

Organ Procurement Under Federal Fire: How Cost-Plus Reimbursement, a 2020 Metric Rewrite, and the DCD Boom Turned Death Determination and Organ Allocation Into Healthcare's Next Compliance Fight Video Preview 🎧 Part I Podcast free on Spotify. 🎧 Part II Podcast episode for paid subscribers only. Also available on Spotify. To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast). Abstract 1. The horror story that got everyone's attention, and the boring machinery underneath it 2. DCD went from 2 percent to half the donor pool in 25 years, and almost nobody outside transplant noticed 3. What the Kentucky review actually found: 351 files, 103 flags, 73 patients, 28 maybes 4. Authorization is not consent, and the gap between them is where NRP lives 5. Follow the money: cost-plus reimbursement, standard acquisition charges, and the missing P and L 6. The 2020 metric rewrite worked exactly as designed, which is the problem 7. Out of sequence allocation, or how the pressure valve became the second scandal 8. What the January 2026 proposed rule does versus what the press release says it does 9. Second order effects for anyone building, investing, or contracting anywhere near this • On August 5, 2026, HHS began decertifying Network for Hope, the OPO serving Kentucky plus parts of Indiana, Ohio, and West Virginia. It is the second OPO ever decertified, after Miami's Life Alliance in September 2025. Before 2025, the number was zero, ever, despite decades of authority to do it. • The proximate trigger was a HRSA review of 351 Network for Hope cases from 2021 to 2024 where donation was authorized but not completed. 103 cases (29.3 percent) showed concerning features, 73 patients had neurological signs incompatible with donation, and at least 28 may not have been dead when procurement started. • The structural context: donation after circulatory death grew from roughly 2 percent of deceased donors in 2000 to 36.1 percent in 2023, 42.9 percent in 2024, and 49.2 percent in 2025, roughly 8,000 donors. • OPOs are cost-reimbursed regional monopolies with no margin pressure, evaluated since 2020 on two outcome measures that only count upside. The first four year recertification cycle under that rule ends in 2026, with tier actions expected January 2027. • Kidney allocation out of sequence went from about 2 percent to over 20 percent between 2019 and 2023 while kidney nonuse climbed from 19 percent to nearly 30 percent. Published work suggests clamping down on it could cost roughly 1,500 kidney transplants a year. • Watch three dockets, not one: the OPO Conditions for Coverage final rule expected late 2026, the FY2027 IPPS organ acquisition cost changes, and whatever HRSA's allocation out of sequence workgroup finally puts out for comment. The horror story that got everyone's attention, and the boring machinery underneath it The version of this story that traveled had a man named T.J. Hoover in it. Kentucky, 2021, overdose, declared brain dead, wheeled toward organ recovery while, by multiple staff accounts, thrashing, crying, and shaking his head. Staff who raised concerns say they were overruled. A congressman called it a horror movie during a July 2025 hearing, which is the kind of line that gets clipped and posted, and it did. The Kentucky AG has been looking at it. That single case has done more to move federal transplant policy than fifteen years of GAO reports and academic complaints combined, which says something a little grim about how health policy actually gets made in this country. Underneath the viral part is a system that is genuinely strange if you sit with it for a minute. There are roughly 55 organ procurement organizations in the United States. Each one is a nonprofit with a federally assigned geographic monopoly, called a donor service area. They are paid on a reasonable cost basis by Medicare, meaning they file a cost report and get reimbursed for what they spent, not for what they produced and not at a negotiated rate. They have no competitors in their territory. Until September 2025 not one had ever been removed, despite CMS having the authority for decades and despite performance spreads between the best and worst that would get a normal vendor fired in a quarter. Public reporting based on 2021 data suggested roughly 42 percent of them were performing at a level that, under the new rules, would eventually put them out. So the interesting question is not whether something bad happened in Kentucky. Something bad clearly happened in Kentucky. The interesting question is what kind of system produces that outcome and then takes four years to do anything about it, and whether the fixes being rolled out now address the machinery or just the headline. For an audience that thinks in incentives, denominators, and reimbursement mechanics, this one is unusually rich, because you can trace a fairly clean causal line from a well intentioned 2020 metric change straight to a patient on a table who was still moving. DCD went from 2 percent to half the donor pool in 25 years, and almost nobody outside transplant noticed Two ways to become a deceased donor. Donation after brain death, where the whole brain including the brain stem has irreversibly stopped and the patient is legally dead while a ventilator keeps the organs perfused. And donation after circulatory death, where the patient has a devastating injury, is not going to recover, is not brain dead, and the family agrees to withdraw life support. The team waits for the heart to stop, observes a no touch period, usually five minutes, declares death by circulatory criteria, and then moves fast because warm ischemia is eating the organs in real time. DBD is clean, or at least conceptually clean. DCD is where the ethics get textured, because the patient is alive right up until the withdrawal, and the entire timeline is scheduled by humans. In 2000, DCD was about 2 percent of deceased donors. By 2023 it was 36.1 percent. By 2024, 42.9 percent. In 2025, roughly 8,000 DCD donors with at least one recovered organ, which is 49.2 percent of all deceased donors with recovered organs. Half. In 25 years an edge case became the modal case, and it happened without anything resembling a public conversation. The technology drove a lot of it. Normothermic regional perfusion, which restores circulation in situ after death is declared, and ex vivo machine perfusion, which keeps organs alive on a pump after recovery, both took hold from roughly 2019 onward. NRP made DCD hearts possible starting in 2019. Machine perfusion approval for livers in 2021 made marginal DCD livers viable. Something like 15 percent of lung transplants in 2024 came from DCD donors. Contemporary DCD donors are older, heavier, more diabetic, more hypertensive, and more likely to have died of anoxia than their predecessors, which is exactly what you would expect from a pool being pushed outward under pressure. Here is a detail that should make any data person twitch. The national registry does not have a field for whether NRP was used. Researchers infer it from the interval between circulatory arrest and aortic cross clamp, because that interval has a bimodal distribution: direct procurement runs short, NRP runs long, so people use cutoffs of 30 or 40 minutes as a proxy. An entire clinical technique, one that involves clamping the vessels to the brain after death is declared, is being studied nationally via timestamp arithmetic. That is not a footnote. That is the state of the evidence base for the fastest growing practice in procurement. What the Kentucky review actually found: 351 files, 103 flags, 73 patients, 28 maybes The numbers are worth stating precisely because the rhetoric around them has been loose in both directions. HRSA examined 351 Network for Hope cases between 2021 and 2024 in which organ donation was authorized but the donation was ultimately not completed. That denominator matters: these are the cases that got called off, which is already a selected sample and arguably the right place to look for near misses. Of those 351, 103 cases (29.3 percent) showed concerning features. 73 patients had neurological signs incompatible with organ donation, meaning levels of consciousness that were high or improving and should have prompted someone to stop much earlier. At least 28 patients may not have been deceased at the time procurement was initiated. The review pointed at poor neurologic assessment, weak coordination between the OPO and the treating clinical team, questionable consent practices, and misclassification of cause of death, particularly in overdose cases. That last one is quietly important. Overdose deaths produce young donors with good organs and neurologic presentations that can look catastrophic and then improve. The pool that expanded the donor supply during the opioid era is the same pool where the death determination is hardest to get right. Multiple patients showed signs of pain or distress while being prepped. Most eventually died hours or days later. Some recovered enough to walk out of the hospital. The federal response ran in stages. July 2025: HHS announces a reform initiative, HRSA directs the OPTN to reopen the case and impose a corrective action plan plus monitoring. Through late 2025 and into 2026: HRSA assesses whether deficiencies were corrected, CMS separately assesses whether the organization still meets Medicare conditions for coverage. May 2026: CMS does a comprehensive review and finds the same categories of problem still surfacing, specifically donor evaluation, adverse event review, and organizational administration. August 5, 2026: decertification begins. Note the shape of that. A year of corrective action, monitoring, and technical assistance, and the follow up survey found the same three buckets. That is either evidence that the organization was unfixable or evidence that corrective action plans are a genre of paperwork rather than an intervention. Both readings have support, and if you have ever watched a nursing home or a hospital work through a CMS plan of correction, you already know which way to bet. The dissent is real and should not be waved off. Network for Hope serves a service area with more than 3,000 people waiting for a lifesaving transplant, and the OPO trade association pointed out that in 2025 more than 46,000 deceased donor organs were transplanted nationally. Decertification is not a scalpel. It is closing a regional utility and hoping the replacement stands up cleanly, in a system where the transition mechanics for handing a DSA to a successor OPO have essentially never been tested at scale. Authorization is not consent, and the gap between them is where NRP lives This is the part that most people outside transplant get wrong, including a fair number of clinicians. When you check the box at the DMV, you are not consenting to a medical procedure. You are making an anatomical gift under your state's version of the Uniform Anatomical Gift Act. That gift takes effect at death and, absent contrary indication from you, cannot be revoked by anyone else, including your family. In 2012 the OPTN formally switched its vocabulary from consent to authorization to reflect exactly this: gift law after death, informed consent before death, two different legal regimes stapled together at the moment of dying. Which creates a seam. DCD requires interventions before death. Lines get placed. Heparin and other medications get administered. These are premortem interventions on a living patient, and they require actual informed consent from the patient's surrogate, not gift law authorization. Meanwhile the anatomical gift itself is governed by a statute that, notably, does not specify how death is to be declared. It just says the gift takes effect on death and leaves the determination to whatever the applicable clinical and legal standards are. Those standards live in the Uniform Determination of Death Act, which is state law, and which the Uniform Law Commission tried to revise a few years back and then shelved when the drafting committee could not get consensus on how to handle brain death criteria. So the country has a rapidly growing procurement pathway sitting on top of a death statute that everyone agrees is imprecise and that nobody can agree how to fix. Now layer NRP on that. Thoracoabdominal NRP involves restarting circulation after death has been declared by circulatory criteria, with the vessels to the head clamped so that brain perfusion does not resume. Proponents argue, with a coherent case, that the cardiac function was irreversible in the sense that mattered, that death already occurred, and that clamping does not cause death because the patient was already dead. Critics point out that if your working definition of death is the permanent cessation of circulation, and you then restore circulation, you have a definitional problem that no amount of committee language resolves. The professional societies have landed roughly on: NRP is ethically defensible, and also families should be told about it explicitly, in plain terms, as part of the DCD conversation. Whether that disclosure actually happens, consistently, in the middle of the worst night of a family's life, at 55 different organizations with 55 different scripts, is an empirical question that nobody has answered nationally. There is no field for it. There is no audit for it. Community engagement work on this keeps circling the same uncomfortable question: is a general DCD authorization sufficient to cover a procedure that some people, for reasons that are cultural and religious and not irrational, would experience as fundamentally different from what they agreed to. The honest answer is that the system has been running the experiment first and asking afterward. Follow the money: cost-plus reimbursement, standard acquisition charges, and the missing P and L Anyone who has done payer contracting will find the OPO economics genuinely disorienting. There is no negotiated rate. There is no price. There is a standard acquisition charge, which is an average cost per usable organ by organ type, calculated by taking the prior year's actual organ acquisition costs from the Medicare cost report and dividing by the actual number of usable organs procured. That charge is what one entity bills another when an organ moves between an OPO and a transplant hospital. It is then reconciled against actual cost at settlement. Chase that arithmetic and you find something interesting. The denominator is usable organs. Fixed costs are enormous: staffed coordinators on call around the clock, hospital development teams, transport, labs. If the organ count goes down, the per organ charge goes up, and Medicare true up covers the gap. If reimbursement exceeds actual cost, the OPO writes a check back. Kidney standard acquisition charges totaled roughly 4.7 billion dollars from 2015 to 2021, averaging about 31,381 dollars per kidney. Medicare is the dominant payer for organ acquisition but only pays its share, roughly half of total claimed acquisition costs in the years where the comparison has been published. The upshot: an OPO does not really have a margin in the way a normal business does. It has a budget and a settlement. There is no financial penalty for a case that gets called off, because the cost of an unusable organ still flows into the pool that Medicare shares in. There is also no financial reward for restraint. What there is, and this is the part that matters, is a strong institutional and reputational reward for volume, plus a workforce whose entire professional identity is built around not letting a gift go to waste. Take away price signals and you do not remove incentives, you just relocate them into metrics and culture. Which is precisely where they went. CMS has started poking at this. The FY2027 IPPS proposed rule published in April 2026 goes after organ acquisition reimbursement in several ways: how standard acquisition charges get reconciled, what counts as an allowable cost, and how overhead gets allocated. CMS's stated view is that providers have been misallocating administrative and general costs using accumulated cost statistics in a way that inflates reimbursement, particularly around organs purchased from OPOs and services purchased under arrangement. There is also a proposal limiting public education costs, which lands directly on the OPOs that have spent real money on stadium sponsorships and awareness campaigns. Those campaigns have historically been booked as allowable cost. Congress has separately been circling the tax exempt status question from the Ways and Means side. Anyone who assumed prior cost report acceptance was protective is going to have a rough 2027. The 2020 metric rewrite worked exactly as designed, which is the problem For decades OPO performance was self reported against denominators the OPOs largely controlled. Everyone knew it was broken. The 2020 Conditions for Coverage final rule fixed it by moving to two objective outcome measures built on external death data: a donation rate and a transplantation rate, both denominated in inpatient deaths among patients 75 or younger with a primary cause of death consistent with organ donation. Benchmarks were set at the lowest rate among the top quartile, reviewed every 12 months, and published. At the end of a four year cycle, OPOs get tiered. Tier 1, the top quartile, auto recertifies. Tier 2 has to compete to keep its service area. Tier 3 gets decertified and cannot compete for anyone else's territory. It worked. Between 2021 and 2023, the number of Tier 1 OPOs doubled from 15 to 30 and Tier 3 dropped from 24 to 10. Over four years, donors rose 31 percent and transplants rose 25 percent. If the goal was to shake loose a system that had been leaving organs on the table, mission accomplished, and the people who fought for that rule against considerable resistance were right. Now read the measures again as an incentive designer would. Numerator: donors recovered, organs transplanted. Denominator: eligible deaths, computed from external data, entirely outside the OPO's control. Nothing in either measure penalizes an aggressive posture on a marginal case. Nothing captures a near miss, a family that felt steamrolled, or a patient who was prepped and then woke up. Those events were not in the scorecard at all until CMS started proposing, in 2026, to codify a definition of unsound medical practices and expand the list of reportable adverse events. Which is a tacit admission that the 2020 scorecard measured production and nothing else, for five years, during the exact period when DCD went from a third of the pool to half. Add the cliff. The first recertification cycle under that rule ends in 2026. CMS is running recertification surveys through late 2026, expects a final rule around then, and plans to commence recertification and decertification proceedings for Tier 2 and Tier 3 OPOs in January 2027. For an organization sitting in Tier 3, the four year run up to that date was an existential sprint to raise a numerator, inside a cost reimbursed structure with no financial brake, using the only elastic supply available, which is DCD, in a donor population heavy on overdose deaths where neurologic prognosis is hardest. Nobody designed that. It is emergent. It is also entirely predictable in retrospect, which is the least useful kind of predictable. Out of sequence allocation, or how the pressure valve became the second scandal Recovery is only half the job. The organ still has to be placed, and placement runs off a ranked match run generated by the allocation algorithm. Sequential offers take time, cold ischemia accumulates, and for a marginal kidney the difference between the eighth call and the eightieth is the difference between a transplant and a discard. So OPOs developed a workaround: bypass the sequence and call a center known to say yes. Formally this is allocation out of sequence, tracked by bypass codes for operational OPO reasons, donor medical urgency, or expedited placement attempts. Then it exploded. Kidney out of sequence allocation went from roughly 2 percent in 2019 to over 20 percent of kidney transplants by the end of 2023. Kidney nonuse over the same window went from 19 percent to nearly 30 percent, which is the tell: the workaround grew alongside the problem it was supposed to solve, not instead of it. By 2023, 15 percent of livers, 8 percent of lungs, and 3 percent of hearts were going out of sequence. Variation across organizations was enormous, from 0 to 43 percent at the OPO level and a nearly identical spread at the transplant center level. On the liver side, a bypass code was used on 3,483 livers over a two and a half year window, of which 2,216, about 69 percent, were transplanted, with donor characteristics similar to standard allocation livers but distinctly different recipients. Same organs, different people, decided by relationship rather than algorithm. There is a defensible version of this. A kidney at high risk of nonuse, called out sequentially for six hours, frequently becomes a kidney that nobody gets. And there is an indefensible version, where a national queue built to be equitable quietly develops a side door and the people with the best relationships walk through it. Published analysis suggests that suppressing out of sequence allocation without building a replacement pathway could mean roughly 1,500 fewer kidneys transplanted, and, more damning, that a regulatory focus on nonuse rates would not even detect that loss. The regulatory response has been fast and a bit blunt. In August 2025 the OPTN sent a memo telling members to follow the match run and stop using unapproved expedited placement variances. Starting November 2025, HRSA began producing monthly allocation compliance reports for every OPO, which is a meaningful shift from annual review to something closer to continuous audit. HHS now cites line skipping safeguards as having protected nearly 300 patients. Meanwhile continuous distribution for kidneys, the long planned successor to the current framework, got paused specifically so the board could work out how the new design would handle nonuse, out of sequence allocation, and expedited placement. In January 2026 a workgroup delivered three draft proposals: an administrative definition of what an offer even is, an expedited protocol for hard to place kidneys, and approaches to compliance. As of that month they were sitting with HRSA's transplant division, HRSA general counsel, and the OPTN board, awaiting public comment. Pause on that first one. In 2026, four decades into the national allocation system, the fix list starts with defining the word offer. Not because anyone is incompetent, but because the practice drifted so far from the policy that the vocabulary stopped mapping to reality. There is also published work showing out of sequence allocation was being underrecognized because of increasing use of free text bypass coding, meaning the measurement of the workaround was itself being routed around. Anyone who has ever tried to run analytics on top of a free text field in a clinical system is nodding right now. What the January 2026 proposed rule does versus what the press release says it does The press release is heavy on sanctity and stewardship. The rule text is narrower and, in a few places, more consequential than the framing suggests. Five things are actually in it. First, the transplantation rate measure excludes pancreata used for islet cell research, plus new recordkeeping to prove that the research is real. That sounds like housekeeping and is not: it closes a numerator loophole where organs routed to research counted toward a production metric. Second, CMS codifies a definition of unsound medical practices, expands the examples of reportable adverse events, and strengthens quality improvement expectations, so that governance and safety failures become independently actionable rather than only visible through the two outcome measures. Third, OPOs must assess and improve their placement performance for medically complex donors, older or sicker, which is a pointed response to the discard problem. Fourth, CMS keeps the tier architecture intact and removes barriers that discouraged high performing OPOs from competing for failing territories. Fifth, and most structurally interesting, it strikes an outdated provision that effectively limited certification eligibility to organizations recertified between 2002 and 2005, with CMS explicitly asking for comment on how to certify genuinely new OPOs. That last one is the sleeper. For a generation the OPO map has been a closed set of incumbents with permanent territories and no entry path. Removing the eligibility limitation plus reducing competition friction plus actually decertifying two organizations plus a January 2027 tier action date adds up to something the sector has never had: contestable geography. Not a market exactly, since the reimbursement is still cost based and the entities are still nonprofits, but a system where a service area can change hands based on performance. Comments closed March 31, 2026, with a final rule expected late in the year and enforcement action 60 days after. What the rule does not do is touch the death determination question. Nothing in it revises how death gets declared, nothing addresses NRP disclosure, nothing creates a registry field for whether NRP was used, and nothing standardizes what a family is told before authorizing DCD. CMS did issue quality and safety guidance in March 2026 that reinforces donor process safeguards and signals tougher surveys, and the expanded adverse event reporting will surface more of these events after the fact. But the actual clinical and legal core, when is a person dead and who decides, remains where it has been: in state statutes, professional guidelines, and 55 sets of local protocols. The federal government is regulating the OPO's homework, not the underlying question the homework is about. Second order effects for anyone building, investing, or contracting anywhere near this A few things follow that are worth pricing in. Compliance and audit tooling around allocation is now a real category. Monthly per OPO allocation compliance reports, a coming administrative definition of an offer, free text bypass coding that defeats analysis, and match run data that cannot even distinguish an offer that was neither accepted nor transplanted: that is a data quality problem with federal enforcement attached to it, which is the classic setup for infrastructure spend. Any organization whose regulatory standing depends on being able to reconstruct why a specific organ went to a specific center at 3 a.m. is about to discover its documentation is worse than it thought. Perfusion is riding two horses. Machine perfusion and NRP are what made the DCD boom possible and are genuinely responsible for thousands of additional transplants, which is the strongest pro growth story in the space. They are also the technologies most exposed if the ethics conversation turns, particularly TA NRP, which depends on a definitional argument about death that has not been settled in law and is being litigated in bioethics journals rather than courts. So far. Anyone modeling that category should carry a real scenario for mandated disclosure requirements, a registry field, or state level restrictions, none of which are far fetched. The medically complex donor mandate in the proposed rule is an underrated demand signal. Requiring OPOs to measure and improve placement of older and sicker donor organs, in a world where nearly 30 percent of recovered kidneys go unused, points straight at viability assessment, predictive acceptance modeling, and anything that shortens the decision loop at a transplant center. The bottleneck is not recovery anymore. It is a surgeon at 2 a.m. deciding whether to take a 78 KDPI kidney with imperfect information and a program level outcome metric hanging over them. The trust variable is the tail risk nobody underwrites. This entire system runs on voluntary registration, and registration is a behavior, not a fixed asset. A steady drumbeat of national coverage about patients waking up on tables is exactly the kind of thing that moves DMV opt in rates, and unlike almost every other variable here, it is not something a rule can fix. The trade association's response to the decertification made this point in a slightly panicked register, arguing that the action should not diminish public confidence in the broader system. That is a legitimate concern and also not how public confidence works. Finally, a note on how the numbers get used, because this audience will notice. HRSA materials say 13 people die each day waiting for a transplant. The CMS proposed rule announcement says 17, and adds that someone joins the list every eight minutes. Both are federal, both are recent, and nobody reconciles them. The waitlist is over 100,000, the deceased donor organ count for 2025 was over 46,000, and the ratio between those two numbers has been the entire policy argument for 40 years. The reforms of the last six years were designed to move it and did, by 25 to 31 percent depending on which measure you pick, which is a genuinely large improvement by the standards of American health policy. The bill for how it was moved is now arriving, in Lexington first, and the interesting part is that the fix and the failure have the same author.

Anthropic Cut Claude Fable 5's Biology Fallbacks By 85 Percent: What The Classifier Rewrite Actually Means For Clinicians, Payers, Pharma R&D, And Anyone Buying Health AI On A Platform Contract

Anthropic Cut Claude Fable 5's Biology Fallbacks By 85 Percent: What The Classifier Rewrite Actually Means For Clinicians, Payers, Pharma R&D, And Anyone Buying Health AI On A Platform Contract Video Preview 🎧 Podcast episode for paid subscribers only. Also available on Spotify. To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast). Abstract 1. What actually shipped on August 7 2. The 85 percent is a ratio with a missing denominator 3. Read the footnote, because the footnote is the story 4. A fallback is not a refusal, and that distinction is a governance problem 5. Dual use has always been a taxonomy fight, and healthcare has been losing it since 1996 6. The clinical middle got the biggest gift and the least ability to check the work 7. Trusted access is the new credential, and credentials are where the rent lives 8. What to build, buy, and bet on over the next twelve months On Aug 7, 2026, Anthropic retuned the biology safety classifier sitting in front of Claude Fable 5 and reported an ~85% drop in biology-related "fallbacks," the silent handoff to Opus 5 that happens when the classifier fires. Key facts and the read: • Fable 5 launched June 9 with nearly all biology and chemistry queries blocked by design. That was a deliberate speed-for-precision trade, and Anthropic said so out loud. • The fix was a rewrite of the classifier's "constitution," expert review, new training data, retraining. Not a threshold slider. A taxonomy rebuild. • Per-surface total fallback reduction: ~67% Claude.ai, ~55% Cowork, ~17% Claude Code, ~7% Claude Platform (the API). • Still blocked: virology, toxicology, molecular design. Which is to say, the actual job of professional biology. • No absolute rates published. No confusion matrix, no PPV, no prevalence. An 85% relative reduction on an undisclosed base. • Comparables: OpenAI's GPT-Rosalind (BixBench 0.751 Pass@1) shipped trusted-access-only from day one. Anthropic shipped broad and gated by classifier. Two different bets on the same risk. • The consumer surface got the most relief, which means the biggest beneficiary is the patient reading a CBC at 11pm, not the medicinal chemist. • For enterprise health AI: silent model substitution breaks reproducibility, breaks validation, and is not currently visible in most DSI transparency reporting. • Strategic take: capability is commoditizing, eligibility is not. The vetting pipeline is the new formulary, and formularies are where margin hides. What actually shipped on August 7 The short version is that Anthropic moved a line. Fable 5 has a smaller model sitting in front of it whose only job is to look at an incoming request and decide whether the topic is one of the ones that could, in the wrong hands, help somebody build something awful. When that little model fires, the request gets punted to Opus 5, which is very capable but which Anthropic assesses as not capable enough to give a bad actor meaningful uplift in biology. Users call this a fallback. Anthropic calls it a fallback. Everybody else calls it "why did Claude suddenly get dumber when I asked about my thyroid panel." On June 9 Fable 5 came out with that classifier cranked wide open, catching almost everything with a nucleotide or a pathogen or a functional group in it. Anthropic was refreshingly blunt about the reasoning: tuning the classifier properly would have delayed general availability by weeks or months, so they shipped with a blunt instrument and ate the false positives. Then there was the export control interruption in mid-June and the restoration on July 1, which is a whole separate essay about how frontier model access is now a trade policy instrument, but set that aside. Over roughly the following six weeks the safeguards team rewrote what Anthropic calls the classifier's constitution, meaning the written rule set the screening model uses to sort safeguarded from allowed. They carved out benign cases in more detail, ran the revisions past internal and external experts, generated fresh training data reflecting the new rules, retrained, and then checked that the thing still caught the stuff it was built to catch. The result, per their own testing, is about an 85% reduction in biology-related fallbacks. Lab result interpretation, symptom questions, general biology education, routine clinical support. All of that flows through to Fable now instead of getting quietly bumped down a tier. What did not change is the part that matters most to anyone in this audience who does biology for a living. Virology, toxicology, and molecular design still fall back. Anthropic said the quiet part plainly: Fable 5 "isn't yet usable for professional biology research and drug development." That is a company shipping a model it describes as capable of outperforming human experts on some complex biological tasks and then, in the same paragraph, telling the people who would benefit most that they cannot use it for their actual work. Not out of stinginess. Out of a threat model that says the same reasoning that designs a better antiviral designs a worse virus, and there is no prompt-level tell that reliably separates the two. The 85 percent is a ratio with a missing denominator Here is where this audience should get twitchy. An 85% relative reduction in fallbacks is a beautiful number and a nearly uninterpretable one, because nobody published the base rate. Run the arithmetic both ways. If the launch classifier fired on 90% of biology-adjacent queries, the new one fires on about 13.5%, which is still one in seven and still annoying but basically livable. If it fired on 30%, the new rate is 4.5% and the problem is functionally solved for general use. If it fired on 99%, which given the "almost all biology queries blocked" framing is not crazy, the new rate is roughly 15%. Three very different products, one identical headline. Anthropic published a conceptual diagram showing a boundary sliding rightward across a spectrum from clearly benign green to dual-use orange to clearly harmful red, and explicitly noted the diagram is qualitative rather than a distribution of actual content volume. Anyone who has ever sat through a diagnostics validation meeting knows what is missing. There is no sensitivity, no specificity, no positive predictive value, no negative predictive value, and no prevalence estimate for genuinely dangerous queries in the wild. A screening test vendor who walked into a health system and said "we cut our recall rate 85%" without disclosing the cancer detection rate would get walked back out to the parking lot. That is roughly the epistemic position everyone is in with this announcement, and it is worth saying that this is not unique to Anthropic. It is the industry norm. OpenAI does not publish a confusion matrix for its bio safeguards either. Google does not. The difference is that Anthropic publishes more of the reasoning and the qualitative structure than anyone else, which paradoxically makes it the easiest company to hold to a higher standard. The other missing number is false negatives. An 85% reduction in false positives is only good news if the false negative rate held. Anthropic says they verified the retrained classifier still generally triggers on harmful and dual-use research content. "Generally" is carrying weight there. Every classifier practitioner knows that moving a decision boundary to reduce type I error without touching type II error requires the underlying model to have actually gotten better at discrimination, not just more permissive. Anthropic claims exactly that, and the rewritten constitution plus retrain is a plausible mechanism for it. But the claim is unfalsifiable from outside. There is no red team report attached, no held-out adversarial benchmark, no jailbreak resistance delta. For a company that has published unusually detailed capability assessments and a whole framework for cyber safeguards jailbreaks, the absence here is noticeable. None of this is a gotcha. It is a request that the field converge on reporting standards, because health systems, IRBs, and pharma governance committees cannot do risk assessment on relative improvements to undisclosed baselines. Give the industry a bio-safeguard equivalent of a STARD checklist and half the arguments go away. Read the footnote, because the footnote is the story The headline says 85%. Footnote one says something far more interesting: total fallbacks, for biology or any other reason, are expected to drop by roughly 67% on Claude.ai, 55% on Cowork, 17% on Claude Code, and 7% on the Claude Platform. Sit with that platform number. Seven percent. The API, which is where every health AI vendor, every EHR-embedded assistant, every ambient documentation product, and every payer pilot actually lives, sees a 7% reduction in total fallback volume from a change that cut biology fallbacks by 85%. There are two readings and both are useful. The first is that biology was never a large share of platform traffic, meaning developers building on Claude are overwhelmingly doing things other than biology, and the fallbacks they were eating were driven by other classifiers entirely. The second, and more interesting, is that scaffolded enterprise prompts do not look like consumer questions. A production clinical documentation pipeline sends the model a structured system prompt, a schema, a transcript, and a tightly scoped instruction. That is a very different distribution than a person typing "is my ferritin of 8 with a normal hemoglobin a problem." The classifier was probably never firing much on the former. Either way, the operational conclusion is the same and it is a little deflating. If a health AI company was hoping this announcement would materially change platform behavior, the answer is that it moves the needle by 7% on total fallbacks and the remaining 93% has other causes. Meanwhile the consumer app got a two-thirds reduction. The people who benefit most from this change are individuals using the chat interface, and the people who benefit least are the companies building regulated products. That inversion deserves more attention than it is getting. Frontier labs are increasingly making their safety posture legible through consumer surfaces, because that is where the complaints are loudest and the screenshots go viral. The reply guy joking that he can finally ask whether two-week-old tofu is edible without being profiled as a bioterrorist is funny and also a perfect summary of the false positive tax. But the entity trying to run a validated clinical workflow does not tweet. It just files a ticket, adds a retry, or switches vendors. A fallback is not a refusal, and that distinction is a governance problem This is the part that should worry anyone with a compliance function. A refusal is loud. The model says no, the user sees it, the log captures it, the eval suite counts it. A fallback is quiet. The request goes to a different model, that model answers, and unless the surface renders a notice, the output looks like a normal answer. It is a silent substitution of the inference engine mid-workflow, triggered by content classification, at a rate the buyer cannot predict. Think about what that does to model validation. A health system that stands up an AI governance committee, runs a local evaluation against a specified model, documents performance on a defined task set, and signs off, has validated a model. If the production traffic silently routes some fraction of requests to a different model based on topic, the validation covers a system that does not exist. The measured performance is a blend of two models weighted by a classifier the health system cannot inspect, whose behavior changes without notice, and whose trigger rate just shifted by 85% in one domain over a weekend. Layer on the transparency expectations already in place for predictive decision support. Developers of certified health IT are supposed to surface source attribute information so users can evaluate whether a decision support intervention is fair, appropriate, valid, effective, and safe. If the DSI documentation names a model and the runtime sometimes uses a different one, the source attributes describe a counterfactual. Nobody drafting those rules was thinking about content-triggered model routing, because in 2023 that was not a thing anyone shipped. It is a thing now. The API case is even sharper because API customers pin model strings. Pinning exists precisely so that behavior is reproducible and billing is predictable. If a request billed against one model identifier is served by a different model because a classifier fired, then the contract, the audit trail, and the reproducibility guarantee are all doing something subtly different than what the buyer assumed. Seven percent of platform traffic is still falling back. For a research pipeline where an evaluator needs to reproduce an output six months later for a regulatory submission, "the classifier felt differently that day" is not an acceptable answer. The fix is not complicated and somebody should ship it. Return a response header or metadata field that names the model that actually served the request and flags whether a safeguard route occurred. Anthropic already exposes plenty of response metadata. Making substitution observable would let enterprise buyers log it, measure it, and account for it in validation, without exposing anything about how the classifier works. Right now the choice is between trusting the routing or building your own detection heuristics, and health IT teams have enough to do. Dual use has always been a taxonomy fight, and healthcare has been losing it since 1996 The philosophical framing around dual use tends to get grand. It is really a boring classification problem that healthcare has been failing to resolve for decades. Dual-use controls are rooted in export-control law that dates back at least to the mid-1990s, and dual-use in the life sciences has long been used to describe legitimate research that could be misapplied to pose a threat to public health and safety.

HRSA's Revised 340B Rebate Pilot Explained: What Claims-Based Rebates on 25 Negotiated Drugs Actually Change for Hospitals, Manufacturers, and TPAs, and Whether It Fixes 340B Transparency

HRSA's Revised 340B Rebate Pilot Explained: What Claims-Based Rebates on 25 Negotiated Drugs Actually Change for Hospitals, Manufacturers, and TPAs, and Whether It Fixes 340B Transparency Video Preview 🎧 Podcast episode for paid subscribers only. Also available on Spotify. To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast). Abstract 1. What HRSA actually did on August 3 and why the second try looks different from the first 2. The mechanics: WAC up front, 45 days to submit, 10 days to pay, and who eats the platform bill 3. Which 25 drugs, which 13 manufacturers, and why less than 5.5 percent of sales is the whole ballgame 4. The float, with actual arithmetic 5. Administrative burden, or how 34,320 dollars a year became a billion depending on who holds the calculator 6. The transparency question: does claims-level data fix what has bugged 340B for a decade 7. Duplicate discounts, nonduplication, and the MFP stacking problem actually driving this 8. No categorical exemptions, and what that means for CAHs, FQHCs, and rural referral centers 9. The legal posture: notice versus rule, the Maine vacatur, and the D.C. Circuit gift 10. The parallel squeeze: ASP minus 33.4 percent, accelerated clawback, and contract pharmacy litigation 11. Who builds the pipes, and why the infrastructure outlives the pilot 12. What to watch between now and April 30, 2028 • HRSA published a revised 340B Rebate Model Pilot Program in the Federal Register on Aug 3, 2026 (FR Doc 2026-15633), effective immediately, as a Notice rather than an NPRM. No new comment period. • Scope: NDC-11s of drugs on the Medicare Drug Price Negotiation Selected Drug Lists for initial price applicability years 2026 and 2027. Roughly 25 products from 13 manufacturers, up from 10 and 8 in the vacated 2025 version. All payers, all indications. • Mechanics: covered entity buys at WAC, submits claims-level data within 45 days of dispense, manufacturer pays or issues a documented denial within 10 days. Rebate equals WAC minus 340B ceiling price on date of dispense. Manufacturer funds the platform. • Timeline: plans due Aug 24, 2026. HRSA approvals by Sept 24. Minimum 90 days notice to covered entities. Go-live Jan 1, 2027. One-year minimum commitment. HRSA evaluation targeted for Apr 30, 2028. • Scale: HRSA pegs the affected products at under 5.5 percent of 2025 340B sales, against a program that grew from 53.7B in 2022 to over 100B in 2025. • Burden fight: HRSA estimates 523.3M dollars annually across 15,249 covered entities (about 34,320 each). AHA says north of 1B for hospitals alone, 150K to 500K+ per hospital. • Legal: the 2025 pilot was enjoined in D. Me. and vacated on remand. Novartis v. Kennedy (D.C. Cir., July 21, 2026) held Section 340B permits rebates but requires Secretarial authorization first, which strengthens HRSA's hand while leaving APA process risk fully live. • Context: CY2027 OPPS proposes 340B reimbursement at ASP minus 33.4 percent and accelerates the conversion factor clawback to 3 percent annually through 2029. What HRSA actually did on August 3 and why the second try looks different from the first Round one of this thing died on the operating table. HRSA put out a rebate pilot notice on July 31, 2025, ran an absurdly compressed application window, approved nine manufacturer models by October 30, blessed Novartis shortly after, and then watched a federal judge in Maine blow the whole thing up on December 29, days before the January 1, 2026 launch. The First Circuit declined to stay the order on January 7. HHS eventually dropped the appeal, the Maine court vacated the pilot and the manufacturer approvals along with it, and by February HRSA was back at square one with a Request for Information instead of a program. Round two arrived July 31, 2026 as an announcement and hit the Federal Register on August 3 as a 64-page Notice. The single most important structural fact about it is that it is not a proposed rule. There is no regulatory text, no proposed CFR amendment, no new comment window, and no comment deadline. It states that it is effective immediately upon publication unless a future notice revises it. HRSA's position is that the deliberative process already happened: a February 17, 2026 RFI, a comment period that closed April 20, and 2,475 total submissions, of which 2,449 were public and 26 were not. A separate Paperwork Reduction Act information collection request ran its own 30-day clock that closed July 15, but that was about burden estimates, not underlying policy. So the agency's theory is roughly: the courts said the first attempt was procedurally sloppy, therefore build a record, respond to the record in writing, and issue the thing as an operative notice. Whether that survives contact with a judge is section nine's problem. What changed on substance is more interesting than the packaging. The 2025 version covered the ten drugs subject to Medicare price negotiation in 2026. The 2026 version adds the 2027 cohort, which roughly doubles the drug count and nearly doubles the manufacturer count. Covered entities got real operational protections that did not exist last year: a minimum submission window, a hard payment deadline, a requirement that denials come with stated reasons and documentation, a rule that manufacturers pay for the technology, and a mandatory 90-day implementation notice. HRSA also added a starting inventory grace period, allowing rebate requests for up to two unreplenished accumulated packages dispensed in the 15 days before go-live, which is a small kindness aimed at the neutral inventory replenishment problem that would otherwise strand product bought before January 1 under a discount regime and dispensed after it under a rebate regime. What did not change is the thing hospitals actually care about, which is that they now front the cash. The mechanics: WAC up front, 45 days to submit, 10 days to pay, and who eats the platform bill Under the traditional model, a covered entity buys a covered outpatient drug at or below the 340B ceiling price. The discount is baked into the invoice. The wholesaler loads 340B pricing, the accumulator does its accounting, and the money never leaves the building. Under the pilot, for a participating manufacturer-drug combination, the covered entity buys through existing distribution channels with wholesale acquisition cost loaded. Full freight. Then it dispenses. Then it has at least 45 calendar days from the date of dispense to submit claims-level data to a manufacturer-designated platform. Then the manufacturer has 10 calendar days from a completed submission to either pay a rebate equal to WAC minus the 340B ceiling price as of the date of dispense, or issue a denial with rationale and specific documentation. If the submission is incomplete, the 10-day clock restarts when the missing data arrives, which is the sort of provision that looks reasonable on paper and becomes a strategic instrument in practice. The platform requirements are the most genuinely provider-friendly part of the notice. The manufacturer pays all costs of the submission platform. The platform has to support all applicable covered entities, provide real-time claims-status reconciliation, limit data collection to what is required, offer technical assistance, and protect PHI and other identifiable information consistent with HIPAA and other applicable law. That last clause is doing a lot of work and is worth staring at, because the entire premise of a claims-based rebate is that a manufacturer, which is not a covered entity and not a business associate in any ordinary sense, is now adjudicating the eligibility of individual dispensing events. Also note the date-of-dispense pricing convention. Ceiling prices reset quarterly. A drug bought in one quarter and dispensed in the next gets rebated at the dispense-quarter ceiling, not the acquisition-quarter ceiling. Pharmacy directors who have spent careers building accumulators around purchase-date logic get to rewire that assumption, and finance gets a new reconciliation variance to explain every quarter. Which 25 drugs, which 13 manufacturers, and why less than 5.5 percent of sales is the whole ballgame The eligible universe is defined by NDC-11 against the CMS Medicare Drug Price Negotiation Selected Drug Lists for initial price applicability years 2026 and 2027, and only during each drug's price applicability period. Critically, the pilot applies across all payers and all indications, not just Medicare. A commercially insured patient's fill of an eligible NDC at a contract pharmacy runs through the rebate path the same as a Part D fill. The 2026 cohort is the familiar first round: Eliquis, Jardiance, Xarelto, Januvia, Farxiga, Entresto, Enbrel, Imbruvica, Stelara, and the NovoLog and Fiasp insulin family. The 2027 cohort brings the semaglutide franchise, Trelegy Ellipta, Xtandi, Pomalyst, Ibrance, Ofev, Linzess, Calquence, Austedo, Breo Ellipta, Tradjenta, Xifaxan, Vraylar, Janumet, and Otezla. Look at that list and notice what is missing. Keytruda is not there. Biktarvy is not there. The oncology and HIV franchises that actually drive 340B economics at large disproportionate share hospitals and Ryan White clinics are almost entirely absent. HRSA's own estimate is that these products represented under 5.5 percent of total 340B sales in 2025, meaning 94.5 percent of the program keeps running on upfront discounts through 2027. This cuts two ways and both of them matter. For manufacturers reading this as a transparency win, it is a very expensive pilot to run on a sliver of volume. Building a claims adjudication platform, staffing a validation function, standing up a denial and appeals workflow, and eating the technology cost, all to gain visibility into a fraction of the transactions where the duplicate discount risk is concentrated, is not obviously net present value positive as a standalone exercise. It is only rational as an option purchase on a much larger future. For covered entities reading this as manageable, the same logic applies in reverse. HRSA explicitly reserved the right to expand to additional drugs and additional price applicability years. IPAY 2028 selections come along on schedule. The pilot is a template, and templates scale. The fixed cost of building the submission and reconciliation muscle is paid once, in 2026 and 2027, on 5.5 percent of volume, and the marginal cost of extending it to 50 percent of volume is small. That is either reassuring or terrifying depending on which side of the invoice a reader sits. The float, with actual arithmetic Everyone says cash flow disruption. Fewer people show the math, so here is a back of the envelope with stated assumptions, which anyone is free to argue with. Start with over 100 billion dollars in 2025 340B purchases, measured at 340B prices. Take 5.5 percent and the pilot cohort is roughly 5.5 billion dollars at ceiling. These are mostly brand products with substantial statutory discounts, so grossing back up to WAC at a blended 45 percent discount puts the WAC-equivalent purchase volume somewhere around 10 billion dollars a year. Call it 8 to 12 billion to be honest about the error bars. Now the timing. The covered entity buys inventory before it dispenses. Then it has up to 45 days to submit. Then the manufacturer has 10 days to pay. Real-world cycle times will not hug the maximums, but they will not hug the minimums either, because month-end batch submission is how these processes actually get run and because the incomplete-submission clock restart is a live risk. Assume 20 to 30 days of inventory carry, 20 to 30 days to submission, and 10 days to payment, and the covered entity is carrying somewhere between 50 and 70 days of WAC exposure at steady state. Sixty days of float on 10 billion dollars of WAC purchases is roughly 1.6 billion dollars of working capital outstanding at any given moment across the covered entity universe. The rebate portion of that, meaning the money that is genuinely a loan to manufacturers rather than ordinary inventory carry, is the WAC minus ceiling piece, call it 45 percent, so roughly 700 to 750 million dollars of interest-free credit extended to pharmaceutical companies at any point in time. At a 5 percent cost of capital, that is 35 to 40 million dollars a year in pure carry cost, sector-wide, on a pilot covering one twentieth of the program. Scale the pilot to the full program and the carry cost lands in the high hundreds of millions annually, which is before anyone talks about denials. Denials are where the model gets spicy. A 2 percent permanent denial rate on 4.5 billion dollars of annual rebate value is 90 million dollars of discount that simply evaporates. A 10 percent temporary denial rate that eventually resolves after resubmission adds weeks of additional float and a whole new labor category. The notice requires rationale and specific documentation for denials, which is a real protection, but it does not create an independent adjudicator, a timeline for appeal resolution, or a penalty for a manufacturer that denies aggressively and settles slowly. The 340B Administrative Dispute Resolution process exists but was not designed for high-volume transactional disputes and would fall over immediately if used that way. The distributional problem is worse than the aggregate. Sector-wide carry cost is an abstraction. A critical access hospital with 25 days cash on hand and a heavy Eliquis and Farxiga dispensing profile through a contract pharmacy network experiences this as a liquidity event, not a line item. Large systems will manage it with revolvers and treasury discipline. Small entities will manage it by not stocking the drug. Administrative burden, or how 34,320 dollars a year became a billion depending on who holds the calculator HRSA's information collection estimate: approximately 523.3 million dollars annually across 15,249 covered entities, averaging about 34,320 per entity, built on an assumption of roughly five hours of additional labor per week. The agency's reasoning is that covered entities already generate, maintain, and transmit substantially all of the required claims data for billing and reimbursement purposes, and that many already use automated systems or third-party administrators, so incremental staffing effects should be modest. AHA's counter is that the annual burden exceeds a billion dollars for 340B hospitals alone, with individual hospital costs ranging from 150,000 to over 500,000 dollars and rising with denial and delay volume. Its central technical objection is that HRSA confused data existing somewhere with data being submission-ready. The elements live across the EHR, the pharmacy dispensing system, the accumulator, the TPA platform, and the claims clearinghouse. Joining them at the transaction level, validating patient definition eligibility and prescriber relationships, confirming that the claim was not billed to Medicaid fee-for-service or a state carve-in arrangement, and then formatting the result for a manufacturer's proprietary portal is not five hours a week of work. It is an integration project followed by an ongoing operations function. Both numbers are advocacy artifacts and both are probably wrong in the same direction. HRSA's estimate is a per-entity average across a population that includes single-site FQHCs and 40-hospital systems, which makes the average nearly meaningless. AHA's per-hospital range is drawn from member self-reporting during an active political fight, which is not a neutral survey instrument. The more useful frame for anyone building a model: this is a fixed cost problem with a small variable component. The integration work, the eligibility logic, the exception handling, and the reconciliation reporting cost roughly the same whether the entity processes 500 eligible claims a month or 50,000. Which means the true burden per claim is wildly regressive against small entities, and the true burden in year one dwarfs the true burden in year three. HRSA's number might be defensible as a steady-state marginal estimate. It is not remotely defensible as a year-one implementation estimate, and the notice's own 90-day advance notice requirement is an implicit admission that implementation is nontrivial. The transparency question: does claims-level data fix what has bugged 340B for a decade Here is the honest answer: partially, and not the part most critics are actually mad about. The decade-old transparency complaint about 340B has three distinct strands that get mashed together in trade press. Strand one is whether covered entities are claiming the discount on claims that do not qualify, meaning patient definition problems, contract pharmacy arrangements where the patient relationship is thin, and duplicate discounts against Medicaid rebates. Strand two is whether the spread between 340B acquisition and third-party reimbursement is being used for charity care and services to underserved populations, or whether it is funding growth capital and specialty pharmacy expansion. Strand three is whether the program's growth from 53.7 billion to over 100 billion dollars in three years reflects genuine safety-net need or arbitrage. Claims-level rebate validation addresses strand one directly and does so well. If a manufacturer sees the NDC, the dispense date, the covered entity identifier, the service location, and enough of the claim to check it against Medicaid Exclusion File status and against its own MFP effectuation records, then duplicate discount detection stops being a retrospective audit exercise and becomes a payment precondition. That is a real change and manufacturers have wanted it for fifteen years. It also gives them something they have never had, which is a defensible transaction-level record for tracking 340B-eligible sales rather than inferring them from wholesaler chargeback data. Strand two it does nothing for. Not a thing. Knowing that a claim was 340B-eligible tells nobody how the margin got spent. That question requires covered entity reporting on the use of program savings, which multiple bills have proposed and none have passed, and which HRSA has no obvious statutory hook to require through a rebate mechanism. Strand three it informs but does not resolve, and only for 5.5 percent of the program. There is also a genuinely uncomfortable second-order point that policy people should sit with. The pilot makes manufacturers the adjudicators of covered entity eligibility on a claim-by-claim basis. Manufacturers are the counterparty whose money is at stake. Under the traditional discount model, the covered entity self-determines eligibility and the manufacturer's remedy is an audit under 340B statutory audit provisions, with HRSA in the middle. Under the rebate model, the manufacturer decides first and the covered entity's remedy is to appeal to the entity that just denied it. That is a meaningful reallocation of adjudicative power that HRSA describes as validation and hospitals describe, not unreasonably, as letting the defendant grade the exam. Duplicate discounts, nonduplication, and the MFP stacking problem actually driving this The drug list is the tell. Every product in the pilot is a Medicare-negotiated drug, and that is not a coincidence or a convenient scoping decision. Under the Inflation Reduction Act, manufacturers must effectuate a maximum fair price for selected drugs. Under Section 340B, they must offer the 340B ceiling price. Statute and CMS guidance both say the two do not stack: a covered entity gets the lower of the two, not MFP applied on top of a 340B price. The operational problem is that in a chargeback world, nobody can tell in real time which unit went where. A wholesaler chargeback tells the manufacturer that a 340B entity bought the product. The Medicare Transaction Facilitator tells the manufacturer that an MFP refund is owed on a dispensed unit. Reconciling those two data streams to prevent double effectuation on the same physical unit, using purchase-side and dispense-side data that were never designed to join, is a genuine mess. A claims-based rebate solves that mess elegantly, from the manufacturer's side, by moving 340B effectuation to the dispense event. Once both MFP and 340B are settled at the claim level, nonduplication is a join on a claim identifier rather than an actuarial estimate. That is why the pilot exists, that is why the drug list looks the way it does, and that is why this is fundamentally an IRA implementation story wearing a 340B costume. Which raises the strategic question nobody at HRSA will answer on the record: if the point is MFP nonduplication, why does the pilot apply across all payers and all indications? A commercial fill of Farxiga at a contract pharmacy has no MFP exposure whatsoever. Applying the rebate model there is not nonduplication policy. It is program integrity policy, or, less charitably, it is scope creep with a nonduplication justification stapled to the front. No categorical exemptions, and what that means for CAHs, FQHCs, and rural referral centers HRSA explicitly rejected carving up the pilot by covered entity category. Not for critical access hospitals, not for rural hospitals, not for federally qualified health centers, not for rural referral centers, sole community hospitals, children's hospitals, freestanding cancer hospitals, or disproportionate share hospitals. The stated rationale is that duplicate discount and nonduplication risk is drug-specific rather than provider-specific, and that observed differences between the average hospital and the average FQHC do not justify excluding entire classes. There is a coherent logic to that, and it is also a striking position for an agency whose press release led with rural and medically underserved communities. What remains is a narrow and entirely discretionary escape hatch. A manufacturer's plan may propose individualized, non-broad exceptions, which must be described to HRSA and communicated to affected providers, with cited examples including entities lacking TPA access, rural hospitals, and health centers. Any accommodation depends on both the manufacturer's willingness to propose it and HRSA's willingness to approve it. It is not a protection. It is a request form. The practical guidance for small entities writes itself. Assume inclusion. Model the working capital hit on the specific NDCs in the two cohorts against actual dispensing volume, not against total 340B spend. Verify that the TPA contract covers claims-level submission to third-party manufacturer platforms and find out what that costs, because TPA pricing was not built for this and will be repriced. Confirm whether the accumulator can distinguish rebate-model NDCs from discount-model NDCs in the same replenishment cycle, because running two pricing regimes through one inventory is the operational core of this whole problem. And press hard for a plan-specific exception in writing during the fall notice period, because once January 1 arrives there is no mechanism to get one. The legal posture: notice versus rule, the Maine vacatur, and the D.C. Circuit gift Two separate legal threads have to be held apart, because trade coverage keeps braiding them. Thread one is whether manufacturers may unilaterally impose rebate models. That thread ended on July 21, 2026, when the D.C. Circuit, in an opinion by Judge Garcia, affirmed the district court in Novartis Pharmaceuticals Corp. v. Kennedy. The holding is precise and consequential: Section 340B permits rebate mechanisms, but the statute vests authority in the Secretary to determine acceptable pricing mechanisms, and the Secretary must affirmatively provide for a rebate mechanism before manufacturers may implement one. Johnson and Johnson, Eli Lilly, Bristol Myers Squibb, and Novartis all lose. Covered entity intervenors, who argued that Section 340B always requires a time-of-purchase discount and never permits rebates, also lose. That second half is the part hospitals should read twice. The strongest available argument against the pilot, the one that says rebates are categorically unlawful, is now foreclosed in the D.C. Circuit. What survives is procedure. Thread two is procedure, and it is very much alive. The 2025 pilot was enjoined by the District of Maine in AHA v. Kennedy on APA grounds, the First Circuit declined to disturb the injunction, HHS abandoned its appeal, and the pilot was vacated and remanded. HRSA's response is to issue the revised program as a Notice, effective immediately, with no proposed regulatory text and no new comment period, resting on a record built from 2,475 RFI submissions. The vulnerability is obvious to anyone who has litigated an APA case. Calling a document a notice does not make it one. If a court concludes that binding covered entities to a new eligibility and payment process, with mandatory submission windows and no opt-out, constitutes a substantive rule affecting rights and obligations, then notice-and-comment rulemaking was required and an RFI does not substitute. HRSA's best defense is that manufacturer participation is voluntary and the notice merely establishes criteria under which the Secretary will exercise preexisting statutory authority, which is a real argument and not a frivolous one. Covered entities' best response is that from their side nothing about it is voluntary, since a covered entity has no general right to remain under upfront discounting for an approved manufacturer-drug combination. AHA has said it is considering all available options, which is lawyer for "the complaint is drafted." Expect a filing in the fall once approved plans publish, because that is when the harm becomes concrete and specific enough to plead. Also note the venue dynamic: the last round was litigated in Maine, not the D.C. Circuit, and nothing about the July 21 opinion binds the First Circuit on the APA question. The parallel squeeze: ASP minus 33.4 percent, accelerated clawback, and contract pharmacy litigation The rebate pilot is not happening in isolation, and anyone modeling 340B hospital economics for 2027 who looks at it alone will be off by an order of magnitude. The CY2027 OPPS proposed rule, released in early July 2026, proposes paying for 340B-acquired drugs at ASP minus 33.4 percent, down from ASP plus 6 percent, based on a hospital acquisition cost survey CMS fielded from January 1 through April 7, 2026. CMS estimates that reduces Original Medicare drug payments by roughly 4.55 billion dollars and beneficiary drug cost sharing by about 1.15 billion in the first year. Statutory budget neutrality requires offsetting that with an increase in nondrug OPPS payments, proposed at roughly 8.44 percent, which redistributes money from 340B hospitals to everyone else. Nonexcepted off-campus provider-based departments would take the same ASP minus 33.4 percent rate without budget neutralization. On top of that, CMS proposes accelerating the clawback of the 2018 through 2022 remedy. The CY2026 final rule set a 0.5 percent annual conversion factor reduction running through 2041 to recover roughly 7.8 billion dollars in nondrug overpayments. The 2027 proposal compresses that to 3 percent annually, finishing by 2029. Same money, one sixth the runway. Meanwhile the contract pharmacy fight grinds on. Roughly 21 states have enacted laws prohibiting manufacturer restrictions on contract pharmacy use. States have generally prevailed in district courts and in two circuits, manufacturers have prevailed in others, and a case remains pending in the Seventh Circuit. The Ninth Circuit meanwhile opened the door wider to False Claims Act theories in 340B enforcement earlier this year. And Congress is circling. A House discussion draft dated June 30, 2026, styled the SECURE 340B Act, and a Senate draft from June 23, 2026, both explore broader structural reform, including questions about whether the program should move toward a rebate architecture generally. Stack it: a Medicare reimbursement cut of nearly 40 percentage points on the drug side, an accelerated clawback on the nondrug side, working capital drag from the rebate pilot, contract pharmacy access under sustained legal pressure, and legislative drafts that would formalize all of it. The rebate pilot is the smallest of these by dollars and the largest by precedent. Who builds the pipes, and why the infrastructure outlives the pilot The most durable observation in the coverage came from Advisory Board's Gina Lohr, who noted that while covered entities are understandably focused on cash flow and administrative burden, the bigger story may be the creation of a new data infrastructure around 340B transactions, giving regulators unprecedented visibility into program utilization and establishing a foundation for oversight well beyond the pilot. That is exactly right and it is where the money is for anyone building or investing in this space. Consider what has to exist by January 1, 2027. Thirteen manufacturers, or however many actually apply by August 24, each need a submission platform meeting HRSA's specifications: universal covered entity support, real-time claims status reconciliation, minimum necessary data collection, technical assistance, and HIPAA-grade protection. No manufacturer is building that in-house in four months. They will buy it, and there are maybe five vendors in the country with the 340B domain knowledge, the claims processing infrastructure, and the compliance posture to deliver it. The existing 340B TPA incumbents have an obvious right to win here, except that most of them are contractually and commercially aligned with covered entities, which creates a conflict that will be interesting to watch get resolved. On the covered entity side, every 340B participant needs a way to identify pilot NDCs at dispense, split them out of the standard replenishment workflow, assemble claims-level submission packages, transmit them to potentially thirteen different manufacturer platforms, track status, reconcile receipts against expected rebate amounts, manage denials and resubmissions, and book all of it correctly. That is a product. Several companies are building it right now. Some of them will be acquired within eighteen months. Then consider what happens if the pilot expands. The fixed cost of that architecture, once paid, extends across the full program. Which means the vendors who win the 2027 contracts are positioned to own the rails for whatever 340B becomes, and the manufacturers funding those platforms are, in effect, subsidizing the construction of an oversight infrastructure they will benefit from and covered entities will be measured by. The strategic irony is thick. Hospitals fought the rebate model and won in court. The manufacturers lost on unilateral implementation and lost again on Secretarial authority. And the entity that emerges with the most durable new capability is neither of them. It is whoever ends up holding the transaction-level dataset. What to watch between now and April 30, 2028 Four dates matter, and none of them is January 1. August 24, 2026 is the manufacturer plan submission deadline, and the first real signal. If all thirteen eligible manufacturers file, the industry has decided the option is worth the premium. If only four or five file, the economics did not pencil at 5.5 percent scale and the pilot becomes a much smaller story. Watch specifically whether the manufacturers with the biggest 340B exposure outside the pilot cohort participate, because they are the ones buying the option on expansion rather than solving a 2027 problem. September 24, 2026 is HRSA's approval deadline, and the approved plans are where the actual policy lives. The notice sets the floor. Each plan sets covered NDCs, platform mechanics, registration requirements, denial procedures, and any individualized exceptions. Covered entities cannot meaningfully plan until those publish, and they publish about 99 days before go-live, which is barely more than the 90-day notice minimum. Whether any plan includes real accommodations for entities without TPA access or with liquidity constraints will say more about how this goes than anything in the Federal Register. The fall litigation window is the third. A complaint filed after plan approval, with named plaintiff hospitals able to plead specific WAC exposure and specific operational impossibility, is a much stronger vehicle than a facial challenge to a notice. If a preliminary injunction lands in November or December, the January 1 date slips again and the pilot enters its third life. April 30, 2028 is when HRSA says it will publish an evaluation using published data to the extent practical and lawful. That hedge is worth noticing. An evaluation constrained by what is practical and lawful to publish, of a program whose entire justification is transparency, is a sentence that deserves to be read aloud slowly. The broader thing to track is scope. HRSA reserved expansion rights explicitly. IPAY 2028 selected drugs arrive on schedule. If the pilot runs a year without operational catastrophe, the argument for extending it to the next negotiation cohort, and then to high-spend specialty products generally, becomes very hard for HRSA to resist and very easy for manufacturers to fund. Under 5.5 percent is a pilot. The same architecture at 40 percent is a different program with the same name, and everything about how the next eighteen months plays out determines which one gets built.

Medicare Implant Pricing and the $25 Billion Blind Spot: Why Devices Escaped IRA-Style Negotiation, What Australia and Japan Already Publish, and What a US Device Price Rule Would Actually Require

Medicare Implant Pricing and the $25 Billion Blind Spot: Why Devices Escaped IRA-Style Negotiation, What Australia and Japan Already Publish, and What a US Device Price Rule Would Actually Require Video Preview 🎧 Part I Podcast free on Spotify. 🎧 Part II Podcast episode for paid subscribers only. Also available on Spotify. To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast). Abstract 1. The tweet, and the one thing in it that's wrong 2. Where the $25 billion actually comes from 3. Medicare doesn't buy implants, and that is the entire problem 4. Why the IRA walked right past devices 5. The gag clause era, and the lawsuits nobody remembers 6. Australia publishes it. Japan publishes it. France publishes it. 7. The claim form is the crime scene 8. What CMS just did in the FY2027 rule, and what it pointedly did not do 9. Bundles as the workaround, and why CJR-X is the real device pricing policy 10. What a device price rule would actually have to say 11. Four ways this goes sideways 12. What to watch • A public back-and-forth about hip implant costs has resurfaced a 15-year-old policy failure: nobody outside a hospital supply chain office can find out what a US hospital pays for an implant. • Medicare spending exposed to implantable device acquisition costs plausibly lands in the $18B to $25B range annually across ortho, spine, cardiac rhythm, structural heart, neuromodulation, and vascular. • Devices escaped IRA-style negotiation for structural reasons, not just lobbying: no ASP reporting, no NDC-equivalent on claims, no rebate program, and payment that is bundled rather than per-unit. • Documented price spreads on identical constructs run 78% to 83% between hospitals, with 90th vs 10th percentile ratios of roughly 2.1x for knees. • The "no country publishes this" claim is false. Australia publishes minimum benefits for 11,000-plus device items. Japan lists roughly 200,000 items across about 1,200 functional categories and repricies them biennially off a transaction-price survey. • The FY2027 IPPS final rule adds a UDI capture measure and nationalizes joint replacement bundles, but still leaves device identifiers off the claim. • A workable rule needs four pieces: identifier on claim, manufacturer net price reporting, a gag clause ban, and outcome data pairing. The tweet, and the one thing in it that's wrong The framing is roughly this: someone with a large audience and a functioning spreadsheet goes looking for what a hip implant actually costs, cannot find it, and says so publicly. The estimate floated is that Medicare "allocates" something like $24 billion for the implants themselves. Comparisons get made to the UK, Canada, Australia. Cue several hundred replies from supply chain directors who know the answer and cannot say it out loud without violating a contract. The reaction from health policy people has been a little smug, which is a mistake, because the underlying observation is correct and has been correct since at least 2012. The one part that is genuinely wrong is the word "allocates." Medicare does not allocate anything for implants. There is no line item. There is no device budget. There is no CMS office that has ever negotiated the price of an acetabular cup. Medicare pays a hospital or an ASC a bundled facility payment for a procedure, and what happens to that money inside the four walls of the facility is, from Medicare's perspective, entirely the facility's business. That distinction matters enormously, because it explains why every instinct imported from the drug pricing fight fails on contact with devices. Drug pricing debates assume a payer paying per unit for an identified product with a reported average price. Device pricing has none of those four things. Not the per-unit payment, not the identified product, not the reported average, and in most settings not even a payer who knows the product was used. So the good news is that the question is the right question. The bad news is that answering it requires rebuilding plumbing that has been broken since the Reagan administration. Where the $25 billion actually comes from Nobody publishes this number, so here is the arithmetic, which anyone with a claims file and a supply contract can reconstruct within a reasonable band. Start with lower extremity joint replacement. Traditional Medicare plus Medicare Advantage runs somewhere in the neighborhood of 800,000 to a million hip and knee replacements a year, now split messily across inpatient, hospital outpatient, and ASC after knees came off the inpatient-only list in 2018 and hips followed in 2020. Implant constructs for a primary case cluster between $4,000 and $6,000 for the big brands, cheaper for commodity and private label, considerably more for revision. Call it $4 billion to $5 billion. Spinal fusion adds another chunk. Medicare volume runs in the low hundreds of thousands per year, and hardware for a multilevel construct lands anywhere from $8,000 to $15,000 once you count screws, rods, interbody cages, and biologics. That is comfortably $2 billion to $3 billion, and the FY2027 rule's creation of new extensive and complex spinal fusion DRGs is a tell that CMS knows the cost distribution has gotten weird. Cardiac rhythm management is where the per-unit numbers get genuinely silly. Pacemakers run a few thousand to eight thousand. Implantable defibrillators, per the last serious federal look at this, had median hospital acquisition prices in the $16,445 to $19,007 range depending on model. Across pacemakers, ICDs, and CRT devices in the Medicare population, another $3 billion is not a stretch. Then structural heart. Transcatheter aortic valve replacement is now the dominant aortic valve intervention in the over-75 population, and the valve itself carries a list price north of $30,000. At current US volumes, and given that this is an almost entirely Medicare-aged procedure, that alone is a $3 billion line. Left atrial appendage occlusion adds another billion-ish. Mitral and tricuspid transcatheter therapies are climbing fast off a smaller base. Add coronary stents, endovascular grafts, spinal cord stimulators at $20,000 to $25,000 a system, shoulder arthroplasty (the fastest-growing arthroplasty segment nobody talks about), trauma hardware, hernia mesh, and roughly four million cataract procedures a year where the intraocular lens is a small unit price times an enormous denominator. The honest range is $18 billion to $25 billion of Medicare-exposed implantable device acquisition cost per year. So the number floating around social media is not crazy. It is in the right order of magnitude, arrived at by the wrong mechanism, which is a pretty good description of most public healthcare cost discourse. For scale: that is larger than what Medicare spent on the first tranche of drugs selected for negotiation under the Inflation Reduction Act. The drug pricing fight got a statute, a negotiation program, a CMS office, and roughly a decade of congressional oxygen. The device pricing fight got a GAO report in 2012 and a lot of nodding. Medicare doesn't buy implants, and that is the entire problem Here is the structure that makes device pricing weird, and it is worth being precise about it because most commentary gets it backwards. The hospital buys the implant. The surgeon picks the implant. The patient receives the implant. Medicare pays the hospital a flat amount for the procedure regardless of which implant was picked. Four parties, and the only one with any price sensitivity is the one with the least control over the decision. This is the classic physician preference item dynamic, and it has survived forty years of attempted reform because it is load-bearing for hospital-surgeon relationships. A hospital that tries to standardize to two vendors is a hospital that risks losing a high-volume arthroplasty surgeon to the joint venture ASC across the street, and with him three hundred cases a year of very profitable outpatient volume. Vendor reps are in the OR. They know the tray. They know the surgeon's preferred offset. Try telling a spine surgeon he is switching pedicle screw systems to save the hospital $2,000 a case and see how that conversation goes. Now run the arithmetic on what that does to margin. Medicare pays roughly $13,500 for an inpatient primary knee under the major joint DRG without major complications. Hospital outpatient lands around $12,200 in Medicare payment on a total procedure cost near $14,100. The ASC rate for a total knee sits near $10,500 in Medicare payment. Against those numbers, a $5,000 implant is 37% of the inpatient payment and roughly 48% of the ASC payment. Compare that to what the surgeon gets. Total knee and total hip both carry 19.11 work RVUs, which after the conversion factor and the 90-day global period works out to something like $1,100 to $1,250 for the operation and three months of follow-up. So the piece of metal and polyethylene costs about four times what the person who implants it gets paid, and the piece of metal has never once been called at 2am about a swollen calf. That ratio is the single most important number in orthopedic economics and almost nobody outside the service line knows it. Why the IRA walked right past devices There is a lazy explanation for why devices got excluded from Medicare drug price negotiation, which is that the device lobby is good at its job. That is true and insufficient. The statute could not have been extended to devices without inventing four things that do not exist. First, an average sales price. Drug manufacturers report ASP quarterly under a specific section of the Social Security Act, net of most discounts and rebates, and Part B pays ASP plus six percent off that reported number. There is no analogous reporting obligation anywhere in device regulation. The FDA regulates device safety and effectiveness and has essentially no interest in what anything costs. CMS has never asked, except in the narrow case of transitional pass-through applications where it does collect invoice cost, which is a fact that will become important later. Second, a rebate infrastructure. The Medicaid Drug Rebate Program forced manufacturers to calculate average manufacturer price and best price, which created decades of pricing data as a byproduct of a completely different policy goal. Devices never had that. There is no device best price. There is no device AMP. There is nothing to build a negotiation formula on top of. Third, a product identifier that appears on a claim. Drugs have had national drug codes since the early 1970s. Every dispensed drug carries one, it flows through the pharmacy claim, and it is the reason you can build a pricing dashboard on a Tuesday afternoon. Devices got unique device identifiers in a 2013 FDA rule, and the identifiers are real and populated in a public database, but they are almost entirely absent from claims. More on that shortly. Fourth, a stable unit to negotiate. A small molecule is the same molecule for twenty years. An orthopedic implant system gets a revised tibial baseplate every eighteen months, a new porous coating, a robotic-assisted version, a cementless option, and a new brand name. Negotiating a three-year price for a product generation that will be commercially obsolete in two is not obviously coherent. This is a real objection, not just industry cover, and it argues for reference pricing or competitive bidding rather than product-by-product negotiation. Put those together and the IRA exclusion looks less like a carve-out and more like an admission that the prerequisites were missing. Which they were. Which they still are. The gag clause era, and the lawsuits nobody remembers The last time the federal government looked hard at this, it found what everyone expected and then the story died. The 2012 GAO work surveyed hospitals, group purchasing organizations, and the federal health systems on what they paid for five categories of implantable device. The results are still the best public data anyone has. One hospital reported paying about $4,500 for a primary total hip construct while another paid about $8,000 for the same construct, a 78% spread. One paid about $5,200 for a primary total knee construct against another at $9,500, an 83% spread. Identical defibrillator models showed hospital-to-hospital differences of $6,844 to $8,723. Spending on procedures involving implantable devices rose from $16.1 billion to $19.8 billion over the five years GAO examined, with orthopedics driving most of the growth. Later work using a larger hospital sample found the 90th percentile hospital paying roughly 2.1 times the 10th percentile for knee implants and about 1.7 times for hips. The same analysis found that hospitals using a joint hospital-physician committee for vendor selection paid 17% less for knees and 23% less for hips, which is the closest thing to a free lunch in this entire domain and which almost nobody implemented. Then there is the payer side, which is where the arbitrage lives. A widely cited analysis of commercial claims found insurers paying roughly $10,605 for knee implants that hospitals acquired for about $5,023, and $11,751 for hips acquired at about $5,620, adding up to several hundred million dollars of spread in a single commercial book over a few years. Under Medicare's bundled facility payment, that markup does not exist as a separate line, which is either a feature or a bug depending on whether you are the hospital. The reason this data is fifteen years old rather than fifteen months old is the confidentiality clause. Device purchase agreements routinely bar hospitals from disclosing net prices, including to benchmarking services. When intermediaries built comparative price databases from hospital-supplied data anyway, at least one large cardiac device manufacturer sued them on trade secret and tortious interference theories. The cases settled before any court ruled on the merits, which was the optimal outcome for the manufacturer: no adverse precedent, and a permanent chill on the entire price benchmarking category. Every general counsel in the space read those complaints and drew the obvious conclusion. So the market has a well-documented 80% price dispersion on identical products, a legal regime that punishes anyone who measures it, and a payment system that gives the buyer no reason to care. That is not a market failure. That is a market functioning exactly as designed by the party with the most to gain. Australia publishes it. Japan publishes it. France publishes it. The claim that no country publishes device prices is the weakest link in the popular version of this argument, and it is worth demolishing because the counterexamples are also the policy roadmap. Australia maintains what is now called the Prescribed List, a legislative instrument specifying the minimum benefit private health insurers must pay for each listed device. It covers more than 11,000 items across cardiac devices, hips, knees, intraocular lenses, human tissue products, and the accessories needed to implant them. Every benefit amount is public. Every billing code is public. The list is updated multiple times a year, with the current version dated mid-2026. It is not a survey or an estimate. It is a published price schedule you can download. Australia also did something more interesting than publishing: it used the published list as a policy instrument. A 2022 agreement between the government and the device industry association set out a methodology for benefit reductions targeting roughly $800 million to $900 million in savings over four years, with the largest tranche landing in July 2022 and benefit cuts applied to about 51% of listed items. That is what happens when you have a published price and the political will to move it. Japan goes further and is the more instructive model for anyone thinking about US mechanics. Japanese national health insurance lists roughly 200,000 device product items grouped into about 1,200 functional categories, each with a reimbursement price. The prices get revised biennially based on a market price survey of what hospitals actually paid, which is the device equivalent of ASP reporting, executed as a government survey rather than a manufacturer obligation. On top of that sits the foreign average price rule, which benchmarks Japanese prices against list prices in the US, UK, Germany, France, and Australia, with outlier trimming to prevent the US number from dragging the average up. If the highest reference price is more than 2.5 times the lowest, it gets dropped entirely. What remains gets capped relative to the mean, and the final Japanese price is held at a multiple of the adjusted average, with a floor preventing any single revision from cutting more than 25%. Read that again with an American eye. Japan has built a device pricing system whose explicit design assumption is that the US price is an outlier to be excluded from any reasonable average. That is not commentary. That is codified in the pricing formula. France maintains a published list of reimbursable products and services with tariffs negotiated by a government pricing committee. Germany does not publish device prices directly but requires a sample of hospitals to submit detailed cost accounting data, including material costs, to the institute that calibrates the DRG weights, which means German DRG rates are built from actual implant cost data rather than charge-to-cost ratios. The UK went at it from the clinical side, with a national improvement program that published implant price variation across trusts and pushed toward standard high-performing implants, backed by a national joint registry and a public rating system that scores implants on documented survivorship. That last piece is the one Americans keep forgetting. The claim form is the crime scene Everything above is downstream of a single mundane fact: the US institutional claim has nowhere to put a device identifier. The unique device identifier system has existed since 2013. Implantable devices carry them. They are registered in a public FDA database with manufacturer, model, brand, and attributes. Certified EHRs have been required to maintain an implantable device list keyed to UDI for a decade. The data exists, it is standardized, and it is sitting in the medical record. It is not on the claim. The institutional claim format adopted under HIPAA has no field for it. Revenue codes for implants exist, but they are cost center buckets, not product identifiers. Outpatient device category codes exist, but they describe categories, not products. The billing committee that maintains the paper claim form approved adding the device identifier years ago, but the electronic transaction standard that actually carries the data has no home for it, and updating the adopted standard is a multi-year regulatory process that nobody has been willing to spend political capital on. The consequence is that CMS pays for roughly a million implant procedures a year and does not know, at the product level, what it bought. Not the brand, not the model, not the manufacturer. When a device is recalled, and they are recalled with some regularity, the agency cannot identify affected beneficiaries from claims. It has to go hospital by hospital. Two of the largest orthopedic recalls of the last two decades, one involving a metal-on-metal hip system and another involving polyethylene inserts that degraded from packaging failures, together affected hundreds of thousands of implanted patients, and the identification process in both cases was a manual scramble. The safety argument for device identifiers on claims is stronger than the pricing argument and has been made repeatedly by former agency heads from both FDA and CMS. The pricing argument rides along behind it. You cannot construct an average sales price for a product you cannot count, and you cannot count what does not appear in any administrative dataset. What CMS just did in the FY2027 rule, and what it pointedly did not do Which brings us to the last few weeks, because the inpatient final rule that dropped in early August contains the most movement on this in years, and also a very deliberate stopping point. CMS finalized a new measure in the hospital interoperability program requiring hospitals to capture the complete unique device identifier for implantable devices in the certified EHR, landing in the public health and clinical data exchange objective for the 2027 reporting period. It is attestation-based at launch, meaning hospitals answer yes or no rather than reporting a numerator and denominator, and a failure to attest costs them the payment update. The agency simultaneously issued a request for information on moving to a performance-based version and on what else UDI capture could support. Read the placement carefully. The measure sits in the interoperability program, tied to EHR capture and public health data exchange. The stated rationale is device surveillance and safety. It does not touch the claim. A hospital can fully comply, capture every UDI perfectly in its EHR, and CMS still learns nothing about which device was implanted in any given Medicare beneficiary, because the data never travels to the agency. That is not an oversight. It is the maximum the agency could do without opening a transaction standard rulemaking and without picking a fight with the device industry over what the identifier would eventually be used for. There has been reporting that CMS is contemplating a proposed rule that would put device identifiers on claim forms. Contemplating. The RFI is the tell that the agency is building a record for that fight. The same rule also proposed eliminating the alternate pathway that let breakthrough-designated devices qualify for new technology add-on payments and outpatient pass-through status without meeting the standard substantial-clinical-improvement test. That is a quiet but real tightening. Pass-through and add-on payments are the only two places in Medicare where device cost is visible as a distinct payment amount, and CMS just made them harder to get. Bundles as the workaround, and why CJR-X is the real device pricing policy If you cannot regulate the price, regulate the buyer's incentive. That has been CMS device strategy for a decade, and it just went nationwide. The final rule confirmed the Comprehensive Care for Joint Replacement Expanded model, mandatory for essentially every hospital paid under the inpatient and outpatient prospective payment systems, covering hip, knee, and now ankle replacements across inpatient and hospital outpatient settings, with 90-day episodes. Start date moved to January 2028, three months later than proposed. Exclusions are narrow: hospitals already in the Transforming Episode Accountability Model, Maryland, critical access hospitals, rural emergency hospitals, and a handful of demonstration sites. That means two mandatory episode models running simultaneously. TEAM launched in January 2026 across roughly 700-plus hospitals in about 188 statistical areas, covering five surgical episode types on 30-day windows, running through the end of 2030, at which point those hospitals roll into the expanded joint replacement model. The predecessor model produced something like $112.7 million in net savings over two performance years while holding quality steady, which is the evidence base CMS cited for going national. Here is why this is device pricing policy in disguise. In an episode model, the implant is no longer a cost the hospital passes through to a fixed DRG. It is a controllable input inside a spend target the hospital gets measured against. The first bundled joint replacement model taught hospitals exactly this lesson, and the industry response was immediate: capitated implant pricing, matrix pricing tied to volume commitments, single-vendor agreements, and the rise of value implant vendors selling perfectly adequate primary constructs at $2,000 to $3,000 against incumbents at $5,000 to $6,000. Make that mandatory for 2,500-plus hospitals and the aggregate purchasing pressure is larger than anything a negotiation statute would have produced in the same timeframe. Every hospital in the country simultaneously acquires a reason to care about implant cost, and the four firms that dominate large joint reconstruction acquire 2,500 counterparties who have all just read the same rule. The catch is that bundles produce private savings, not public information. A hospital that negotiates its knee construct from $5,500 to $3,200 keeps the difference until the benchmark rebases. Nothing about that transaction becomes visible to CMS, to researchers, or to the next hospital. The dispersion problem does not get solved. It gets redistributed toward whoever has the best supply chain team, which in practice means the large systems that already had the best prices. What a device price rule would actually have to say Strip away the advocacy and a workable rule has four components. Three of them are boring and one of them is the whole fight. The first is identification. Device identifiers have to appear on institutional claims for implantable devices, which requires CMS to move on the adopted transaction standard and give hospitals a real implementation window, probably three years, with the identifier populated automatically from the EHR implantable device list rather than hand-keyed by a coder. Without this, nothing else in the list is possible. With it, every other element becomes an analytics problem rather than a policy problem. The second is price reporting. Manufacturers report average net selling price by device identifier on a quarterly lag, net of rebates, discounts, purchasing organization administrative fees, and any other consideration. This is not novel. It is the drug ASP mechanism with a different identifier, and the statutory language could be adapted in an afternoon. Alternatively, follow Japan and run it as a periodic transaction price survey of purchasers, which shifts the reporting burden to hospitals but captures the actual paid price rather than the manufacturer's calculation of it. The invoice cost collection that already happens for pass-through applications proves the agency can handle the data. The third is the gag clause ban. Federal law already prohibits contractual terms that prevent group health plans from accessing their own claims data, enacted as part of a broader transparency package a few years back. Extending the same principle to device purchase agreements is a one-paragraph provision that would immediately reopen the benchmarking market that litigation shut down. This is the cheapest, highest-yield item on the list and it requires no new infrastructure whatsoever. The fourth is outcome pairing, and this is the one that separates a serious proposal from a press release. Publishing prices without survivorship data creates pressure toward the cheapest construct, and orthopedics has a long, expensive history of cheap-and-novel implants that failed at five years and generated billion-dollar settlements. The UK solved this by pairing its price transparency push with a national registry and a public rating system that grades implants on documented years of survivorship evidence. The US has a joint replacement registry with millions of procedures in it, but participation is voluntary, it does not link to price, and it does not link to claims. Device identifiers on claims would connect all three datasets automatically, which is the strongest argument for the first item on this list and the one that should be leading the pitch. Four ways this goes sideways Transparency in concentrated markets is not automatically pro-competitive, and anyone selling it as a sure thing is not being careful. The canonical counterexample is a Danish antitrust experiment in ready-mix concrete, where the regulator published firm-level transaction prices in a concentrated market and prices converged upward rather than downward, because publication gave each firm perfect visibility into whether rivals were discounting. Large joint reconstruction is dominated by four firms. Publishing counterparty-identified transaction prices in that structure is a plausible way to make the low-price hospitals pay more rather than making the high-price hospitals pay less. The mitigation is to publish distributions with a lag rather than named transactions, which preserves the benchmarking value while removing the enforcement mechanism for tacit coordination. The second risk is that price becomes the only visible attribute. See above regarding cheap implants and five-year revision curves. A revision costs Medicare two to three times a primary and costs the patient a year. The third is the surgeon relationship. Every hospital that has tried aggressive implant standardization has discovered that the surgeon is mobile and the hospital is not. Mandatory bundles help here by giving the hospital a regulatory reason rather than a margin reason, but the ASC alternative is a real and growing pressure valve, and the ambulatory surgery industry has already told CMS in writing that pulling ASCs into episode accountability would slow the migration of procedures out of higher-cost settings. They are not entirely wrong. The fourth is innovation, which is the argument the device industry will lead with and which deserves a fair hearing even from skeptics. The US is where new device categories get paid for first, and the margin from US pricing funds the iteration. Japan's own experience is instructive here in an uncomfortable direction: as its foreign average price rule tightened and the innovation premium on new device categories compressed from roughly 10% to about 3% over a decade, the ratio of Japanese prices to the foreign average fell below parity, and the debate there is now whether the country is systematically underpaying for new technology. Whatever the US builds should have a real premium pathway for genuine clinical improvement, which makes the simultaneous tightening of the breakthrough device pathway in the FY2027 rule look a little like closing the innovation door while pretending to shop for a lock. What to watch Three things over the next eighteen months. Whether CMS actually proposes device identifiers on claims. The RFI in the FY2027 rule is the strongest signal in a decade, and the safety framing gives the agency political cover that the pricing framing never would. If a proposed rule appears, the comment file will be the most interesting document in health policy that year, and the device industry's comments will tell you precisely how much the current opacity is worth to them. Whether the mandatory bundle rollout produces visible implant price compression. Hospitals have seventeen months before the expanded joint replacement model starts. Supply chain teams are already renegotiating. The public evidence will show up first in device manufacturer earnings calls as pricing pressure commentary in large joint reconstruction, and then in the DRG relative weight recalibrations two or three years later as the cost report data catches up. Whether the transparency bills moving through Congress pick up a device provision. Several hospital and insurer transparency bills advanced through committee this year, all of them extending the existing machine-readable file framework for services and negotiated rates. None of them touch device acquisition cost. Adding a gag clause ban would cost nothing, score as savings, and be nearly impossible to oppose publicly. Somebody should hand that amendment to a staffer. The larger point is that the person asking what a hip implant costs was not being naive. He was pointing at the last major category of US healthcare spending with no price data at all, in a system that has spent fifteen years building elaborate transparency machinery for hospital charges and negotiated rates while leaving 40% of the cost of a joint replacement completely dark. The drug pricing debate got a statute. The device pricing debate has a request for information and a checkbox in an interoperability program. That is progress, technically. So is a glacier.

Populist Instincts, Institutional Podium: What The Kennedy Versus Dana Bash CNN Interview Reveals About Vaccine Messaging, HHS Credibility, And The Future Of Federal Public Health Communication

Populist Instincts, Institutional Podium: What The Kennedy Versus Dana Bash CNN Interview Reveals About Vaccine Messaging, HHS Credibility, And The Future Of Federal Public Health Communication Video Preview 🎧 Podcast episode for paid subscribers only. Also available on Spotify. To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast). Abstract A live Sunday-show exchange between the HHS Secretary and a CNN anchor became a proxy fight over something larger than any single vaccine claim: whether the communication style that built a movement can survive contact with running a roughly $1.7T department. Core arguments: • Populist health messaging optimizes for perceived candor and adversarial framing. The Secretary role is graded on predictability, coverage plumbing, and institutional trust. Different objective functions. • "Show me the placebo-controlled trial" is unbeatable in an 11-second clip and close to useless operationally, because the record it dismisses is the same record underwriting payer coverage, VFC eligibility, and liability protection. • The tenure has a measurable record: ACIP dissolved and reconstituted, roughly $500M in mRNA countermeasure contracts canceled, HHS headcount cut from about 82K toward about 62K, 28 divisions folded into 15, a CDC director ousted about 4 weeks after confirmation, and the worst measles year since the early 1990s. • The institutional response was not rebuttal but routing around: AHIP coverage pledges running through 2026, regional state compacts issuing their own recommendations, and AAP, ACOG, and IDSA publishing independent schedules. • Net result is de facto federalization of the immunization schedule, a real cost borne by payers, providers, and manufacturers regardless of who "won" the segment. • For investors and operators: model ACIP as a variable input, not a constant. Price schedule risk into vaccine, diagnostics, and pediatric benefit assumptions. 1. Why a nine-minute cable segment counted as a policy event 2. What a populist optimizes for, and why the job grades on something else 3. The placebo problem, or how "show me the study" became a closing argument 4. Receipts from the tenure, since vibes are not a dataset 5. Why the institution cannot punch back in real time 6. The credibility ledger, as measured by people who move money 7. Fifty schedules, none of which anyone asked for 8. The counterargument, which deserves better than a strawman 9. What competent looks like from here, assuming anyone wants it Why a nine-minute cable segment counted as a policy event The Sunday show is a genre with rules older than most of its audience. Anchor plays a clip, guest explains the clip, anchor reads the guest's own words back at them, everyone breaks for commercial having generated roughly zero new information. It is health policy kabuki with better lighting. The Kennedy and Bash exchange broke the format not because either party did anything unusual for their profession, but because the two professions were running incompatible software on the same stage. Bash was doing accountability journalism, which needs a fixed referent: you said this, the record says that, reconcile them. Kennedy was doing movement communication, which needs no fixed referent, because the movement's founding proposition is that the record itself is the thing under suspicion. When one side treats the official record as the scoreboard and the other treats it as the opposing team, there is no debate. There are two monologues sharing a split screen. That mismatch is the actual story and the reason the clip traveled. Health policy people saw an interview. Comms people saw a structural fault line. The Secretary sits atop a department that outspends the GDP of most countries, employs tens of thousands of scientists and regulators, and issues recommendations that flow straight into insurance contracts, pharmacy protocols, school entry rules, and standing orders in pediatric practice. That office has never been a good venue for adversarial performance, not because bureaucrats are fragile, but because its output is consumed as a stable input by systems that cannot rebalance every ninety days. A senator can freelance. A cabinet secretary who freelances is functionally issuing guidance, whether or not the Federal Register agrees. There is also an audience-capture dynamic worth naming plainly. The interview was not aimed at CNN viewers. Nobody watching live was going to change their mind, and the Secretary's team surely knew it. The target was the clip, and the clip's audience is a distributed network that has spent a decade being told legacy media flattens dissent. In that frame, an interview that reads as hostile is not a loss, it is a deliverable. Getting interrupted is the point. Any campaign operative could have explained this. What is less appreciated is how differently the tactic behaves when the person deploying it also controls the Vaccines for Children program. What a populist optimizes for, and why the job grades on something else Populist health communication has a coherent internal logic, and pretending otherwise is how the establishment lost the argument in the first place. It optimizes for perceived candor over precision, for naming villains over describing systems, and for validating the listener's experience rather than correcting it. Those are not dumb choices. They are extremely effective choices for assembling a coalition of people who have spent twenty years being told their kid's food is fine, their chronic pain is idiopathic, and their questions about the schedule make them dangerous. The chronic disease critique underneath the movement is not fringe. Something like six in ten American adults carry at least one chronic condition, the pediatric obesity curve is genuinely grim, ultraprocessed calories dominate the diet, and the federal nutrition apparatus has produced decades of guidance ranging from useless to actively counterproductive. Anyone who has watched a dietary guidelines cycle up close knows the capture critique is not paranoid. The problem is the job description. A Secretary gets graded on a different variable set, and the grading is done by systems rather than voters. Predictability comes first. Plans build actuarial assumptions on multi-year cycles, and a recommendation that flips mid-plan-year does not merely annoy people, it strands money. Coverage plumbing comes second. The preventive services provisions of the Affordable Care Act tie no-cost-sharing immunization coverage directly to advisory committee recommendations, which makes those votes a coverage trigger with roughly the force of contract. Liability architecture is third. The vaccine injury compensation system, its excise tax funding, and the manufacturer protections that make a low-margin, high-litigation-exposure category commercially viable at all sit downstream of the same machinery. Fourth is the boring one, procurement. The federal government buys vaccines for roughly half of American children through a program whose eligibility list is defined by those votes. Change the vote, change who gets the shot for free. None of this is visible from a studio, which is why the two modes cannot be reconciled with better media training. A populist optimizes for the feeling that someone is finally telling the truth. A Secretary optimizes for systems that read every sentence as a parameter change. When one person attempts both, the systems absorb the volatility and the movement gets the applause, and the invoice for the first arrives long after the second has scrolled past. The placebo problem, or how "show me the study" became a closing argument The most durable rhetorical move in the entire vaccine fight is the demand for a saline-controlled trial. It is unbeatable on live television and close to meaningless in practice, and explaining why takes about ninety seconds of methodology that no cable segment will ever allow. That asymmetry is the whole game. The actual situation is this. Many childhood vaccines were licensed against an active comparator or an adjuvant-matched control rather than saline, and the reason is not a cover-up. Once a product has demonstrated efficacy against a serious disease, randomizing children to receive nothing stops being a study design and becomes an ethics violation, and no review board in the developed world will approve it. Same principle that prevents anyone from running a saline-controlled trial of insulin. Meanwhile the evidence that does exist is enormous and unusually well suited to catching rare harms: linked administrative datasets covering tens of millions of person-years, self-controlled case series designs that use each patient as their own control and therefore wipe out confounding by indication, national registry studies from Denmark and Finland following entire birth cohorts for decades, and active surveillance that has in fact caught real signals. The intussusception problem that pulled a licensed rotavirus product off the market was found this way. So was the myocarditis signal in young men after mRNA doses. The system detects things, which is the actual evidence that the safety record is not a rubber stamp. Try compressing any of that into eleven seconds while being interrupted. It cannot be done. "Where is the placebo trial" fits in one breath. "Self-controlled case series over linked national registry data" fits in a graduate seminar. The populist move wins on format alone, and the anchor's only available counter is an appeal to authority, which is precisely the move the movement exists to reject. Watching a skilled interviewer walk into that loop is instructive. Every escalation toward consensus reinforces the frame that consensus is the subject of the story. The cost lands somewhere else entirely. When that argument comes from the podium rather than a podcast chair, it reads as a signal about future regulatory posture. Manufacturers hear licensure standards being reopened retroactively. Investors hear a discount rate. Health departments hear a green light for parents who were already wavering. The statement does not have to be wrong to be expensive. It only has to come from the person who controls the schedule. Receipts from the tenure, since vibes are not a dataset Grading a cabinet secretary on one interview is lazy, so consider the record the interview was arguing about. The advisory committee on immunization practices was dissolved in full during the first summer, all seventeen sitting members removed at once, replaced by a much smaller panel whose composition drew immediate objection from the senator whose vote had made confirmation possible. That panel moved against thimerosal-containing influenza formulations, a fight settled scientifically in the early two thousands. Later meetings shifted the hepatitis B birth dose away from a universal recommendation toward individual decision making, and pushed pediatric Covid vaccination into shared clinical decision territory. Earlier still, the recommendation for healthy children and healthy pregnant women had been altered by video announcement, without the committee voting at all, which is the part that mattered structurally. The process was not overruled. It was bypassed. Elsewhere the department canceled roughly half a billion dollars across more than twenty mRNA countermeasure projects, a decision defended on platform-specific grounds that virologists found unpersuasive and that quietly rewrote the pandemic preparedness posture. The reorganization folded twenty-eight operating divisions into fifteen, cut regional offices in half, and drove headcount from around eighty-two thousand toward sixty-two thousand through a mix of layoffs and departures. The confirmed CDC director lasted about four weeks before being pushed out, followed by the resignation of several of the agency's most senior career scientists, one of whom left with a public letter that read like a resignation from the profession rather than the post. A flagship children's health report shipped with citations to studies that did not exist, which is the kind of error that gets a graduate student expelled. And a White House event linked a common analgesic in pregnancy to autism on evidence that the obstetric societies rejected within hours. Against that, the year's hardest number was measles. The country logged its worst year since the early nineteen nineties, with three deaths, the first in a decade, concentrated in an undervaccinated West Texas community. Officials promoted vitamin A alongside vaccination, and a Lubbock children's hospital ended up treating pediatric hypervitaminosis A, which is a genuinely novel way to generate iatrogenic harm during an outbreak of a preventable disease. Elimination status went from a settled fact to an open question. Why the institution cannot punch back in real time The most underrated feature of the whole confrontation is that the institution had no available response. Federal health agencies communicate through a process built for accuracy at the expense of speed: clearance chains, science review, legal review, press office coordination, and an aversion to unforced error that borders on pathological. That machinery produces excellent documents in about four weeks. The clip cycle resolves in about four hours. By the time a careful correction exists, the argument has moved on, and the correction itself gets read as a defensive institutional maneuver rather than information. Worse, the agencies report to the person they would be correcting. Career staff who once might have pushed back through a friendly reporter now do it anonymously or not at all, and the anonymous version is trivially reframed as the deep state doing what deep states do. Professional societies stepped into the gap, but a specialty society statement carries no coercive force. It can influence clinician behavior, but it cannot rewrite federal forms, repeal statute, or compel a payer to honor a no-cost-sharing promise if the underlying rule changes. That is why the public fight kept sliding away from evidence and toward process. Process is where the state still has leverage; evidence is where the movement likes to fight. The credibility ledger, as measured by people who move money The most important outside signal came from insurers. AHIP and several large carriers publicly said they would continue covering vaccines that had already been recommended under existing standards through at least the end of 2026, a move designed to prevent immediate benefit disruption and reassure employers that plan-year assumptions would not be blown up by a committee vote. Regional state compacts did something similar by issuing their own immunization recommendations and implementation guidance, effectively creating a parallel standard set for clinicians and public health departments. Pediatricians and obstetricians responded in kind: the AAP, ACOG, and IDSA published independent schedules rather than waiting for federal harmonization. That is the real institutional story. Coverage did not wait for the federal government to clarify itself. It routed around uncertainty. In practical terms that means the schedule has become more federalized and less centralized at the same time: federal law still matters, but the functional operating standard is increasingly whatever a payer will honor, a hospital system will stock, and a specialty society will defend in public. That is a high-friction way to run a country, and it is expensive in ways that do not show up in a single headline. It also means manufacturers, providers, and employers have to treat ACIP as a variable input. The people who move capital already are. Fifty schedules, none of which anyone asked for The counterargument, which deserves better than a strawman, is that the previous regime was not some frictionless golden age. Vaccine policy has always been partly political, advisory committees have always had members with prior commitments, and federal public health messaging has its own long history of overclaiming. There is also a legitimate argument that more transparency about evidence quality, uncertainty, and dissent would have helped. None of that is false. What is different now is not that conflict exists. It is that conflict has been made the operating system. Once the Secretary's podium becomes a venue for permanently contesting the legitimacy of the schedule, every downstream actor is forced to choose between formal federal guidance and functional continuity. That is not reform. It is fragmentation with a moral vocabulary. What competent looks like from here, assuming anyone wants it Competence here would look boring, which is part of why it is so hard to sell. It would mean separating movement communication from operational authority, restoring the norm that advisory processes are advisory rather than theater, and treating immunization recommendations as infrastructure instead of as culture-war content. It would mean acknowledging uncertainty when it exists without turning uncertainty into a universal solvent. And it would mean understanding that the public does not need every question answered on television; it needs institutions that can keep the lights on while the questions are answered elsewhere. The Kennedy-Bash interview will be remembered as a confrontation, but its deeper significance is simpler: it showed that the country now has two vocabularies for the same system, and they are no longer translating in real time.

The CMS autism toolkit is a build spec in disguise: where the 421 percent ABA spending curve creates real openings in utilization management, credentialing, documentation, and payment integrity

The CMS autism toolkit is a build spec in disguise: where the 421 percent ABA spending curve creates real openings in utilization management, credentialing, documentation, and payment integrity Video Preview 🎧 Part I Podcast free on Spotify. 🎧 Part II Podcast episode for paid subscribers only. Also available on Spotify. To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast). Abstract 1. What CMS actually shipped and why the 421 percent number matters 2. The ABA market as it exists today, warts included 3. Where the money leaks and who gets paid to plug it 4. Utilization management that clinicians will not hate 5. Credentialing and supervision is a data problem wearing a compliance costume 6. Documentation, session capture, and the AI note trap 7. Payment integrity, SIU tooling, and the very slow, very rich state buyer 8. Diagnosis, wait lists, and the front door of the funnel 9. What this does to platform economics and the PE roll-ups 10. Outcomes measurement as the only moat that survives a rate cut 11. Underwriting these bets without getting flattened by the next policy swing CMS released the State Medicaid and CHIP Applied Behavior Analysis Toolkit in early August 2026. Headline data point: ABA spend in Medicaid and CHIP grew **421%** from **2017** to **2022** against **66%** growth in children served. The toolkit is explicitly not a new federal requirement and explicitly does not touch EPSDT. What it does is hand more than 50 state agencies and territories a shared vocabulary for medical necessity, treatment intensity, provider qualification, and fraud detection in a benefit that went from rounding error to line item in about six years. Key takeaways for builders and investors: • The spend/utilization gap (421 vs 66) is mostly hours per child and rate creep, not more children. That is a UM problem, and UM problems are software problems. • The rendering provider gap is the structural flaw: technicians deliver the care, BCBAs own the NPI on the claim. Every integrity product eventually points here. • EVV for in-home ABA is the obvious next mandate. Nobody has built the good version yet. • State agencies can buy program integrity tooling at 90/10 and 75/25 federal match. That changes the ROI math on a slow sales cycle. • CMS-0057 prior auth APIs land Jan 2027. ABA is the highest-friction PA category nobody built for. • Evidence base is thin enough that outcomes measurement is the only defensible position once rates compress. What CMS actually shipped and why the 421 percent number matters Read the press release twice and the second read is more interesting than the first. CMS says multiple times, in different ways, that the toolkit is not a restriction, does not create federal requirements, does not reduce EPSDT obligations, and does not endorse a particular flavor of ABA. That is the sound of an agency lawyer standing behind the policy shop with a hand on their shoulder. EPSDT is a statutory entitlement, states have lost in court repeatedly when they tried to hard cap medically necessary services for children, and CMS knows it. So the toolkit is guidance, checklists, and suggested practices, which in Medicaid-speak means states can adopt any of it, all of it, or none of it, and the ones with budget problems will adopt most of it by Q2 next year. The 421 percent against 66 percent is the whole ballgame. Spend grew roughly six times faster than the population being served. There are only three ways that happens: rates went up, hours per child went up, or the billing got creative. Rates did go up in a handful of states but nowhere near six-fold, and several states cut. So it is hours and it is billing. A child authorized at 12 hours a week and a child authorized at 38 hours a week generate wildly different annual costs, roughly $30k versus close to $100k depending on the state rate, and the clinical literature supporting the difference between those two is thinner than anyone in the industry likes to say out loud in a room with an actuary. That gap is where the companies get built. Not in the politics of it, which are exhausting and will be litigated for years, but in the plumbing that lets a state or a plan tell the difference between the two children without generating 400 appeals a month. The ABA market as it exists today, warts included Autism prevalence in the CDC’s surveillance network is now about **1 in 36** eight-year-olds in the most recent published cycle, up from **1 in 44** two cycles prior and roughly **1 in 150** twenty years ago. Roughly half of those children touch Medicaid at some point. Every state has to cover medically necessary ABA for children under EPSDT, a posture CMS made explicit in a 2014 informational bulletin, and every state also has a commercial autism mandate on the books. Demand is not the constraint. Supply is. The workforce is a two-tier structure that nobody would design on purpose. A little over **65,000** Board Certified Behavior Analysts nationally, and something in the neighborhood of **200,000** Registered Behavior Technicians. The BCBA writes the treatment plan, does the assessment, modifies protocols, and supervises. The RBT does the actual hours, sitting on the floor with a child running discrete trials, usually for **$20 to $24 an hour** with turnover that clinics quietly admit runs **65 to 75 percent** a year. A staffing model with three-quarters annual churn in the delivery role is not a clinical model, it is a call center with better intentions. The billing runs through a small set of CPT codes that most people outside the space cannot keep straight. 97151 for assessment, 97153 for the technician-delivered direct treatment in 15 minute units, 97155 when the behavior analyst modifies protocol, 97156 for caregiver guidance, plus the group codes. 97153 is the volume driver and it is where the entire economic argument lives. State Medicaid rates on it typically run **$12 to $16 a unit**, call it **$48 to $64 an hour**, with real spread from the low $30s to north of $80 depending on the state and whether there is a managed care markup. The default treatment intensity is where things get uncomfortable. The 40-hour-a-week comprehensive model traces back to a 1987 study with 19 children in the treatment arm. That study did something important for the field, and it also became a load-bearing wall for a multibillion dollar benefit in a way its author probably did not anticipate. The Council of Autism Service Providers now distinguishes focused treatment at roughly 10 to 25 hours from comprehensive at 26 to 40, which is a reasonable framework, and which a lot of providers treat as a menu where every child orders the tasting menu. Where the money leaks and who gets paid to plug it CMS mentions investigations, prosecutions, and convictions related to kickbacks and harm to children. That is not abstract. The pattern in the enforcement actions is consistent enough to be a product spec. Recruiters paid per head to bring children into clinics. Diagnoses generated by clinicians with financial relationships to the treatment provider. Hours billed for sessions that did not occur or occurred with a family member watching TV nearby. Telehealth sessions billed as in-home direct treatment at the higher rate. Technicians whose credentials lapsed months ago still generating claims. And the classic: a single BCBA NPI on claims totaling more than 24 hours in a calendar day, which is either fraud or a genuinely remarkable person. That last one is the structural flaw and it deserves its own paragraph. In most states the RBT cannot enroll as a Medicaid provider, so the claim carries the supervising BCBA’s NPI as the rendering provider even though the BCBA was not in the room. The claim line literally does not know who delivered the care. Every downstream integrity question, every supervision ratio check, every no-show detection scheme, every attribution model for outcomes, runs into this wall. Some states have started requiring technician-level identifiers, either as an ordering/referring relationship or through a modifier convention, and it is a mess of state-by-state variation. Building the resolution layer for that is a real company. Not a fraud dashboard, a registry and attribution service that maps technician identity, active credential status, supervision relationship, and session-level presence to the claim line, and does it across the credential registry, the state enrollment file, the practice management system, and the payer’s claim history. Boring. Necessary. Very hard to rip out once it is in. Utilization management that clinicians will not hate The toolkit’s language about treatment intensity reflecting each child’s clinical needs rather than standardized service models is a polite way of saying stop authorizing 40 hours by default. The states that have tried the blunt version have gotten mixed results. Indiana is the cautionary tale everyone cites. It watched ABA spend go from a rounding error in the late 2010s to a few hundred million a year, then in 2024 cut the 97153 rate to roughly **$68 an hour** and layered on a 30-hour weekly cap with a multi-year limit on comprehensive treatment. Providers left, families sued, and the state ended up in a fight it partly created by not having a defensible clinical process in the first place. The lesson for builders is that hard caps are legally fragile under EPSDT and politically radioactive. Individualized medical necessity determination is neither. It is just expensive to do well, which is the definition of a software opportunity. What the product looks like: an authorization engine that ingests the assessment instruments, the treatment plan, the goal set, and the prior authorization period’s actual progress data, and produces a defensible hour recommendation with the clinical rationale attached. Tiering logic that distinguishes focused from comprehensive on something other than the provider’s request. Automatic step-down triggers when goals are mastered. Escalation paths that route the 8 percent of genuinely complex cases to a human reviewer instead of the 92 percent that are template renewals. The timing hook is CMS-0057. Impacted payers, including Medicaid fee-for-service and managed care plans, have to stand up FHIR-based prior authorization APIs by Jan 1, 2027, with decision timelines of 72 hours expedited and seven calendar days standard already phasing in. Everybody built for radiology and ortho. Almost nobody built for behavioral health, and ABA is arguably the highest-friction PA category in Medicaid: long authorization periods, huge dollar values, clinical documentation that lives in a specialty practice management system that does not speak FHIR, and a renewal cadence that hits every six months forever. Whoever ships the ABA-native DTR and PAS implementation gets to be the default. Credentialing and supervision is a data problem wearing a compliance costume CMS calls out helping states ensure ABA providers are qualified and properly supervised. Two separate problems hiding in one sentence. Qualification is an enrollment and screening question. States already have authority under the provider screening regulations to assign risk categories, require fingerprint-based background checks, and unannounced site visits. Several states have moved or are moving ABA providers into the high risk category, which triggers the full screening battery. That creates work: revalidation cycles, ownership disclosure, site visit logistics, license and certification monitoring against the certifying board’s registry, and the tedious reality that a mid-size ABA group with 40 clinics across six states is managing thousands of individual credential records with a shared inbox and a spreadsheet named final_v4. Supervision is the more interesting problem because it is continuous rather than periodic. The certifying board requires that a percentage of technician hours be supervised, and states layer their own requirements on top. Nobody verifies this at the claim level in anything close to real time. A supervision monitoring product that reconciles technician direct-service hours against documented supervision contacts, flags ratio breaches before they become recoupments, and produces the audit packet on demand would sell to providers as risk insurance and to plans as an integrity control. Same underlying data model, two buyers, very different pitches. The adjacent play is audit defense as a service. When states get aggressive with post-payment review, and they will, the recoupment demands land on providers who kept documentation in a system that was designed for scheduling, not for surviving a statistical extrapolation. There is a services business and eventually a software business in extrapolation challenges, sampling methodology disputes, and appeal workflow. Unglamorous, high retention, and it grows exactly when the market gets scary. Documentation, session capture, and the AI note trap Here is a prediction that seems safe: electronic visit verification comes for in-home ABA within three years. The federal EVV mandate under the 21st Century Cures Act already covers personal care and home health services, states built the infrastructure, and the same fraud pattern that motivated it (services delivered in a home with no independent witness, billed in 15 minute increments) applies to a benefit that just grew 421 percent. The toolkit does not say EVV. The toolkit does not have to say EVV. Some state Medicaid director reading the fraud section is going to connect those dots by lunch. The version of that product that works is not the one that treats a clinician like a delivery driver. Existing EVV implementations are widely hated for exactly that reason. The version that works captures the session data the clinician was going to collect anyway, trial-by-trial