Thoughts on Healthcare Markets and Tech (@thoughtson_tech) - Twitter Profili

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

On Feb 13 2026, DOGE dropped a 10.32 GB Medicaid claims file with zero friction. 227M rows. 1.8M provider NPIs. 7 years of billing data. No application form. No data use agreement. Just a link. Within days, public Jupyter notebooks had mapped autism billing anomalies in MN to active FBI cases. DMEPOS storefronts in TX and FL lit up on heatmaps. Telehealth phantom visit patterns surfaced in multiple states. The internet became a fraud funnel. Federal validation followed fast. By May 2026, DOJ added 15 trial attorneys for Medicaid fraud and ran a 15-defendant MN takedown. The distributed analysis had not replaced investigators - it had expanded what investigators could prioritize. The commercial question: $1.5T in commercial spend. $30-60b/year in estimated recoverable improper payments. No equivalent public data. No cross-payer signal infrastructure. The legal scaffolding to build one exists. Nobody has built it yet. Subscribe to onhealthcare.tech for free and paid articles, podcasts, and more. For a further deep dive on the topic from today's video teaser, see the podcast and article link onhealthcare.tech/p/how-doge-ope…

English

0

1

38

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

Worked with a health system last year where a hospitalist built a custom discharge summary tool in about four days using Cursor. No eng team, no IT ticket, no vendor. Just a doc with a problem and a free weekend (and, honestly, a lot of coffee). That jump you're describing is exactly the mechanism I've been tracking from the payer and vendor side. When a physician can go from tinkerer to builder in a weekend, the prior auth workflow tool that a vendor spent eighteen months encoding stops looking like infrastructure and starts looking like a weekend project. But here's where it cuts deeper than most people are saying. The hospitals most able to act on this aren't the small systems. A system above $2 billion in revenue has enough volume and enough workflow complexity to justify even a $300K internal build. And at that scale, the math on buying a bolt-on Epic ecosystem tool versus building it internally has already flipped. The vendor risk isn't slow erosion. It's that the highest-value, highest-volume customers leave first, and what remains is a long tail of smaller buyers who couldn't build anyway. That's a tough cap table story for a Series A company whose thesis was basically "we encoded the rules so you don't have to." The physician-developer tier you're naming is the supply side of that shift becoming real. onhealthcare.tech/p/the-free-lun…

English

0

1

2

27

Praneet Mylavarapu, MD@PMylavarapuMD·2d

A taxonomy that's been useful for thinking about where physicians sit: Skeptic → Consumer → Power user → Tinkerer → Physician-Developer The jump from Tinkerer to Physician-Developer used take years. As of early 2026, it requires a weekend and a clear problem.

English

2

1

3

1K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

Frontier AI labs are the right peer group for valuing Isomorphic right now, but the "SaaS in the future" frame skips a structural wrinkle worth sitting with. The biobuck economics in Iso's Novartis and Lilly deals (roughly $37.5M and $45M upfront against $1.2-1.7B in milestone payments) are exactly what deferred-revenue SaaS looks like before the contracts convert. The headline numbers get cited as partner conviction; the actual cash transfer is contingent on clinical proof points that don't exist yet. The atom-world moat argument holds, but the capital structure tells you who is actually betting on it. No pharma corporate venture on Iso's cap table despite $2.7B raised is a signal, not an accident. That keeps M&A options open while the UK sovereign equity stake (which the market seems to be treating as a footnote) could constrain which buyers are even eligible down the road. Full breakdown of the capital stack and what it means for the category: onhealthcare.tech/p/isomorphic-l…

Ryan Bethencourt@RyanBethencourt

EVERYONE is talking about AI x Bio/Science The venture money today is still in foundational models but ultimately the long term moats will be in the world of atoms (physical products). Many science foundation models will look like SaaS businesses in the future but massive asymmetric value will be in building in the hard spaces. Building the biotech companies that have to push against the broken regulatory system, raise money for studies, risk everything on new molecules, cell therapies or other innovations. Most Pharma's todays were born in the 1800's, only a handful of real founder CEO's have been able to survive and grow in the BioPharma space. The new moats are massive amounts of real patient data (companies like Tempus AI), conviction and experience with real technologies like Abscellera and their dedication to antibody technology, surviving through booms and busts like Twist as they scale DNA manufacturing. Pharma today exists because of daring entrepreneurs who were able to scale their businesses (literally door to door selling anti parasitic candies like Pfizer) and persist because of the massive regulatory and cash moats necessary to get FDA approval. There are many other paths but most of these AI x Bio foundational model companies aren't what will scale, it'll be the founders at the coal face throwing dice and finding n = 1 paths to fund studies and approval of novel drugs AND everything will be AI supported as table stakes (and Quantum when it comes online). Biology is real nanotechnology, it's alien, programmable tech that we're just starting to understand how to harness with AI to literally reprogram reality, not just our bodies but scarcity and our planet.

English

0

43

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

Frontier AI labs are the right peer group for valuing Isomorphic right now, but the "SaaS in the future" frame skips a structural wrinkle worth sitting with. The biobuck economics in Iso's Novartis and Lilly deals (roughly $37.5M and $45M upfront against $1.2-1.7B in milestone payments) are exactly what deferred-revenue SaaS looks like before the contracts convert. The headline numbers get cited as partner conviction; the actual cash transfer is contingent on clinical proof points that don't exist yet. The atom-world moat argument holds, but the capital structure tells you who is actually betting on it. No pharma corporate venture on Iso's cap table despite $2.7B raised is a signal, not an accident. That keeps M&A options open while the UK sovereign equity stake (which the market seems to be treating as a footnote) could constrain which buyers are even eligible down the road. Full breakdown of the capital stack and what it means for the category: onhealthcare.tech/p/isomorphic-l…

English

0

8

Ryan Bethencourt@RyanBethencourt·5d

EVERYONE is talking about AI x Bio/Science The venture money today is still in foundational models but ultimately the long term moats will be in the world of atoms (physical products). Many science foundation models will look like SaaS businesses in the future but massive asymmetric value will be in building in the hard spaces. Building the biotech companies that have to push against the broken regulatory system, raise money for studies, risk everything on new molecules, cell therapies or other innovations. Most Pharma's todays were born in the 1800's, only a handful of real founder CEO's have been able to survive and grow in the BioPharma space. The new moats are massive amounts of real patient data (companies like Tempus AI), conviction and experience with real technologies like Abscellera and their dedication to antibody technology, surviving through booms and busts like Twist as they scale DNA manufacturing. Pharma today exists because of daring entrepreneurs who were able to scale their businesses (literally door to door selling anti parasitic candies like Pfizer) and persist because of the massive regulatory and cash moats necessary to get FDA approval. There are many other paths but most of these AI x Bio foundational model companies aren't what will scale, it'll be the founders at the coal face throwing dice and finding n = 1 paths to fund studies and approval of novel drugs AND everything will be AI supported as table stakes (and Quantum when it comes online). Biology is real nanotechnology, it's alien, programmable tech that we're just starting to understand how to harness with AI to literally reprogram reality, not just our bodies but scarcity and our planet.

English

8

2

19

3.2K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The 45.3% hitting ≥30% TWL at 12 mg is already eating into sleeve gastrectomy territory, but you're right that lean mass data is the real swing variable. If DEXA confirms better body composition, that's when payers and ICER models get truly destabilized, not just repriced at the margin. onhealthcare.tech/p/eli-lillys-t…

English

0

10

Greg O'Gallagher@gregogallagher·1d

If Reta shows better lean muscle retention in the June 6 Dexa scan phase 3 data I will consider switching to Reta (assuming I don’t mind the higher RHR) and get Pharma grade (not research lol)

English

23

0

44

31.7K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The heart rate signal is real and the glucagon receptor piece is almost certainly the explanation. Glucagon agonism increases cardiac output directly, separate from the sympathetic activation you'd see with caloric restriction, so the chronotropic effect you're describing isn't incidental noise, it's a predictable pharmacodynamic consequence of the third agonist that tirzepatide doesn't carry. What gets me about your experience specifically is the baseline mismatch. The TRIUMPH-1 cohort entered at 112.7 kg and BMI 40.0, heavier than SURMOUNT-1, and the cardiometabolic readouts they're reporting (systolic BP reduction, hsCRP improvement) look clean in aggregate. But aggregate cardiometabolic benefit in a heavy, inflamed cohort can statistically swamp an HR elevation that would be clinically significant for a leaner or already-cardiovascularly-stressed switcher. You're not the target demographic the trial was powered around, and that gap matters. The outcomes trial question is where this gets structural. As I wrote in onhealthcare.tech/p/eli-lillys-t…, Medicare coverage and payer authorization pathways for retatrutide are going to hinge partly on unresolved outcomes data requirements, and the cardiovascular load question you're raising is exactly the kind of signal that regulators will want answered in a dedicated CVOT before broad coverage unlocks. Three months of sustained HR at 84 from a 62 baseline is not nothing. That's worth tracking formally with whoever prescribed this.

English

0

50

Peptide Confessions@pepfessions·1d

Switched from tirzepatide to retatrutide expecting magic. Got the same weight loss but my resting heart rate climbed from 62 to 84 and stayed there for 3 months. Reta is not tirz plus. It's a different beast and nobody talks about the cardiovascular load.

English

80

13

419

107.5K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

What happens when the distributor leverage on generics gets compared to the formulary leverage PBMs hold over brand manufacturers — are those two pressure points additive, or does one dominate? The PBM side of this is where the ceiling gets set for something like Cost Plus. The Big Three PBMs control roughly 90% of U.S. prescription claims, and the mechanism isn't just pricing pressure. It's the threat of portfolio-wide formulary deprioritization across a manufacturer's entire drug lineup if they supply Cuban at competitive brand pricing. A manufacturer weighing one brand drug deal against potential formulary exclusion across their whole portfolio isn't doing a simple margin calculation. But the distributor concentration you're describing operates one layer upstream of that, at the physical supply chain level rather than the benefit design level. And the question your example raises is whether those two chokepoints compound each other in ways that make the brand drug market structurally unreformable from the retail end, which is exactly the ceiling Cost Plus keeps running into. That's the part TrumpRx's generic expansion doesn't touch. Coupons and cash pricing can route around PBM gatekeeping on generics. They can't route around formulary threats on brands without regulatory intervention at the PBM architecture level. onhealthcare.tech/p/cuban-joins-…

English

0

1

12

Mark Cuban@mcuban·1d

Let me give you an example of where there is no gov intervention, and the impact on brand drug pricing. When a brand manufacturer sells a drug to one of the big 3 drug distributors that control more than 90 pct of their market, those multi hundred billion dollar distributors DONT negotiate the lowest price they can get. They literally pay retail price. Then, in exchange for paying promptly, and providing some data, they get a discount of a whopping 5 pct. For a $600 drug, their net cost is $570 For obvious reasons, that distributor can’t sell to your local pharmacy for less than $570. So when you go to buy that drug, and have no insurance, or a deductible of more than $600, that’s why you pay the full $600. The question is “why would multi hundred billion dollar distributors only negotiate a 5% discount on brand drugs?” I asked this very question to several CEOs of brand drugs companies First you have to know that the pharma companies don’t keep that full $570. Because they pay rebates and fees to the big insurance company PBMs , they end up netting about 50% , or $300 in this example I asked them why they didn’t sell to the big distributors at a little more than their net price, which would allow them to make more money. And it would also allow the distributors to sell to pharmacies at say $350 (so the distributors make more money ), and the pharmacies could sell to the uninsured and those during their deductible phase for $375. Meaning more patients could benefit from their drugs. This doesn’t mean every patient could afford their meds, but it means that more could. Saving $225 is not nothing. The CEOs each told me that they would like to, but can’t. Why? Because the ins company PBMs have told them that if they did this , they would reduce their position on their formularies. Which could cost them billions of dollars across all their drugs. None of this is against the law. It’s become standard industry practice. Until we break up these conglomerates , it will only get worse.

English

106

101

628

265.1K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

What happens when the distributor leverage on generics gets compared to the formulary leverage PBMs hold over brand manufacturers — are those two pressure points additive, or does one dominate? The PBM side of this is where the ceiling gets set for something like Cost Plus. The Big Three PBMs control roughly 90% of U.S. prescription claims, and the mechanism isn't just pricing pressure. It's the threat of portfolio-wide formulary deprioritization across a manufacturer's entire drug lineup if they supply Cuban at competitive brand pricing. A manufacturer weighing one brand drug deal against potential formulary exclusion across their whole portfolio isn't doing a simple margin calculation. But the distributor concentration you're describing operates one layer upstream of that, at the physical supply chain level rather than the benefit design level. And the question your example raises is whether those two chokepoints compound each other in ways that make the brand drug market structurally unreformable from the retail end, which is exactly the ceiling Cost Plus keeps running into. That's the part TrumpRx's generic expansion doesn't touch. Coupons and cash pricing can route around PBM gatekeeping on generics. They can't route around formulary threats on brands without regulatory intervention at the PBM architecture level. onhealthcare.tech/p/cuban-joins-…

Mark Cuban@mcuban

Let me give you an example of where there is no gov intervention, and the impact on brand drug pricing. When a brand manufacturer sells a drug to one of the big 3 drug distributors that control more than 90 pct of their market, those multi hundred billion dollar distributors DONT negotiate the lowest price they can get. They literally pay retail price. Then, in exchange for paying promptly, and providing some data, they get a discount of a whopping 5 pct. For a $600 drug, their net cost is $570 For obvious reasons, that distributor can’t sell to your local pharmacy for less than $570. So when you go to buy that drug, and have no insurance, or a deductible of more than $600, that’s why you pay the full $600. The question is “why would multi hundred billion dollar distributors only negotiate a 5% discount on brand drugs?” I asked this very question to several CEOs of brand drugs companies First you have to know that the pharma companies don’t keep that full $570. Because they pay rebates and fees to the big insurance company PBMs , they end up netting about 50% , or $300 in this example I asked them why they didn’t sell to the big distributors at a little more than their net price, which would allow them to make more money. And it would also allow the distributors to sell to pharmacies at say $350 (so the distributors make more money ), and the pharmacies could sell to the uninsured and those during their deductible phase for $375. Meaning more patients could benefit from their drugs. This doesn’t mean every patient could afford their meds, but it means that more could. Saving $225 is not nothing. The CEOs each told me that they would like to, but can’t. Why? Because the ins company PBMs have told them that if they did this , they would reduce their position on their formularies. Which could cost them billions of dollars across all their drugs. None of this is against the law. It’s become standard industry practice. Until we break up these conglomerates , it will only get worse.

English

0

24

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The question this raises: does it matter *why* people stop, or only that they do? With GLP-1s it matters a lot. Two-thirds of weight regained within a year of stopping isn't a willpower story, it's a biology story. But 68 percent of patients quit within 12 months anyway, and when I modeled what that costs a 100,000-member plan, the wasted spend hit 4.7 million dollars annually, just from early stops. That's the number payers are starting to price. When discontinuation becomes a line item, the question shifts from "so what" to "who pays to fix it and how." onhealthcare.tech/p/the-glp-1-go…

English

0

1

13

Crémieux@cremieuxrecueil·22h

Most of those who make it their New Year's resolution to lose weight give up within a few months. So what? Yes, if you stop using GLP-1s, dieting, exercising, going to dance classes... then you'll be doing something else.

The Wall Street Journal@WSJ

While nearly 18% of U.S. adults have taken a GLP-1 drug for weight loss or to treat a chronic condition, about half of people will stop taking it within a year. Often, they don’t understand what is likely to come next. 🔗: on.wsj.com/4dCkbia

English

7

10

204

162K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The "detects good vs bad cells" framing is doing a lot of work here, and it's worth slowing down on what that actually means mechanistically. Rodent VEGF modulation data is real, but VEGF signaling is notoriously context-dependent. The same pathway drives wound healing and tumor angiogenesis, and compounds that modulate it in gastric tissue in rats have repeatedly failed to translate cleanly into human oncology applications. The selective cytotoxicity claim needs human pharmacokinetic data, receptor binding characterization, and dose-response curves across cell types before it carries the weight you're putting on it. But the deeper problem is categorical. When I analyzed the wellness peptide market for a piece on engineered signaling biology, the issue wasn't whether BPC-157 has interesting rodent data. It does. The issue is that mouse models showing tumor suppression or VEGF disruption get laundered into safety and efficacy claims for human use, and the gap between those two things is precisely where patients get hurt. Lutathera's NETTER-1 data and PSMA-617's VISION trial exist because those compounds went through the receptor binding and dose-response work that characterizes legitimate peptide drugs. BPC-157 has not done that. And the cancer safety question specifically cannot be resolved with rodent lifespan models alone. The absence of a clear carcinogenic signal in mice is not the same thing as a clean human safety profile, especially for a compound sold through mail-order "research use only" vendors with no manufacturing quality controls. onhealthcare.tech/p/the-peptide-…

English

0

18

Morph@doctormorphh·23h

The fear of BPC-157 inducing cancer growth is unjustified. BPC-157 can detect whether a cell is 'good' or 'bad' and adapt its mechanism to disrupt VEGF signaling in cancer cells, decrease muscle loss from cancer and prolong lifespan of mice in cancer models. BPC-157 kills cancer. r: pubmed.ncbi.nlm.nih.gov/41155565/ and read: pubmed.ncbi.nlm.nih.gov/40573323/

Morph@doctormorphh

BPC-157 can potentiate the effects GH has in tendons by increasing and potentiating the GH receptor. BPC-157 also enhances the JAK2 pathway, which is needed for GH to produce IGF-1 making BPC-157 more interesting than it already is.

English

8

34

351

32.1K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

Walked through this exact dynamic when analyzing ACA marketplace enrollees at 200-400% FPL. Post-subsidy premiums drop under $100/month, which looks like a policy win, but the silver plan deductible sitting at $3,000 to $8,000 means the insurance card in someone's wallet is functionally decorative for most of the year. The hospital-as-subprime-lender framing you're building toward is accurate, and it connects to something the deductible structure creates upstream: people skip the $150 primary care visit for acid reflux management or a thyroid adjustment (neither of which is covered under ACA first-dollar preventive requirements, by the way), the condition compounds, and they end up in the ED where the hospital can capture the insurance payment while financing the patient's share through a payment plan that is, yes, a subprime instrument by any functional definition. The insurer gets to price premiums against a population that self-rations primary care until it becomes acute, the hospital gets paid (eventually, partially), and the patient holds debt at terms they had no real power to negotiate. That's not a bug in the system, it's load-bearing architecture. What makes this hard to fix is that the access failure happens in the gap between premium affordability and deductible exposure, a gap that neither regulators nor most policy proposals have been willing to look at directly. Direct Primary Care subscriptions at $50-150/month could theoretically bridge it, but the IRS hasn't definitively ruled those HSA-eligible, so even that workaround is legally ambiguous. onhealthcare.tech/p/the-direct-p…

Mark Cuban@mcuban

Explainer: If you need care and you can't afford your deductible, you will have to borrow money to pay for it. Typically the healthcare provider will loan you the money. Why ? They want to get your insurance company's money. Now, by definition, the hospital is a subprime lender. Not only is the patient in a huge financial hole, across their similarly situation patients, so is the hospital. They know they won't collect however much your deductible is. And that's just for your in-network deductible, nor does it account for your family max out of pocket I don't know the % of bankruptcies this causes What I do know is that the loss of hundreds of billions of dollars for patients and hospitals, starts with the plans offered by insurance companies. They know damn well when someone picks a plan and they can't afford the deductible. And to them, it's not a bad thing. If you can't afford your deductible, the chances they pay anything from your premiums, go way way down So patients go broke or can't afford care. Hospitals have huge uncollected debt, so they make it up elsewhere And then they aggregate those amounts, among others, and use them to get payments from state and federal programs. See how all that works together ?

English

0

52

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

Walked through this exact dynamic when analyzing ACA marketplace enrollees at 200-400% FPL. Post-subsidy premiums drop under $100/month, which looks like a policy win, but the silver plan deductible sitting at $3,000 to $8,000 means the insurance card in someone's wallet is functionally decorative for most of the year. The hospital-as-subprime-lender framing you're building toward is accurate, and it connects to something the deductible structure creates upstream: people skip the $150 primary care visit for acid reflux management or a thyroid adjustment (neither of which is covered under ACA first-dollar preventive requirements, by the way), the condition compounds, and they end up in the ED where the hospital can capture the insurance payment while financing the patient's share through a payment plan that is, yes, a subprime instrument by any functional definition. The insurer gets to price premiums against a population that self-rations primary care until it becomes acute, the hospital gets paid (eventually, partially), and the patient holds debt at terms they had no real power to negotiate. That's not a bug in the system, it's load-bearing architecture. What makes this hard to fix is that the access failure happens in the gap between premium affordability and deductible exposure, a gap that neither regulators nor most policy proposals have been willing to look at directly. Direct Primary Care subscriptions at $50-150/month could theoretically bridge it, but the IRS hasn't definitively ruled those HSA-eligible, so even that workaround is legally ambiguous. onhealthcare.tech/p/the-direct-p…

English

0

17

Mark Cuban@mcuban·20h

Explainer: If you need care and you can't afford your deductible, you will have to borrow money to pay for it. Typically the healthcare provider will loan you the money. Why ? They want to get your insurance company's money. Now, by definition, the hospital is a subprime lender. Not only is the patient in a huge financial hole, across their similarly situation patients, so is the hospital. They know they won't collect however much your deductible is. And that's just for your in-network deductible, nor does it account for your family max out of pocket I don't know the % of bankruptcies this causes What I do know is that the loss of hundreds of billions of dollars for patients and hospitals, starts with the plans offered by insurance companies. They know damn well when someone picks a plan and they can't afford the deductible. And to them, it's not a bad thing. If you can't afford your deductible, the chances they pay anything from your premiums, go way way down So patients go broke or can't afford care. Hospitals have huge uncollected debt, so they make it up elsewhere And then they aggregate those amounts, among others, and use them to get payments from state and federal programs. See how all that works together ?

Anish Koka, MD@anish_koka

“Did medical bills single-handedly account for more bankruptcies than anything else? No. This is an exaggerated half-remembering of a series of studies, authored by (among others) Elizabeth Warren, that were themselves exorbitant exaggerations.” - @asymmetricinfo

English

79

74

531

255.3K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The hearing aid result is real progress, but it also clarifies exactly where the bottleneck sits. A 12dB improvement and 75-95% preference rates in epilepsy patients means the neural decoding works, the problem is that these decoders were calibrated under controlled conditions and will drift as the electrode-tissue interface degrades, typically within 1-2 years from glial scarring. The clinical win and the engineering liability are the same system. And that gap is where I'd push back on reading this as a straight path to commercialization. The attended speech decoding here is impressive, but who owns the adaptive recalibration layer when the decoder drifts in a real patient six months post-implant? That's the question Neuralink and Merge Labs haven't answered publicly, and it's showing up in their job postings. I mapped this specific dynamic when writing about picks-and-shovels opportunities in the BCI infrastructure layer, the case that the most defensible startup position right now isn't building a competing platform but solving one named bottleneck before the giants can do it internally. Decoder drift and biocompatible electrode longevity are the two I'd watch most closely, because a result like this one just put a clock on... onhealthcare.tech/p/brain-comput…

English

0

11

Medscape@Medscape·20h

A brain-controlled hearing system, tested in epilepsy patients, accurately identified and amplified attended speech, improving perception by 12 decibels. Participants preferred this system in 75%-95% of trials, marking a significant step toward practical application. mdsc.pe/4ulHX9j

English

1

5

8

1.6K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The $37,824 number lands harder when you understand who's actually absorbing it. For the 100 million Americans in self-funded plans, that cost sits directly on the employer's books, not an insurer's. Every dollar spent is a dollar the employer is watching, which is exactly why employers stopped being passive benefit purchasers a long time ago. That shift is what I've been writing about. When employers carry the direct financial risk, they start making clinical decisions, not just financing ones. Boeing negotiating bundled orthopedic pricing that determines where employees get surgery. Amazon running therapeutic substitution protocols where internal clinical teams intervene in prescribing. These aren't benefit design tweaks. They're healthcare decisions made by entities whose primary interest is your productivity, not your health. The structural cost problem is what's driving all of it. At $13,800 average cost per employee per year, a mid-sized employer with 500 people is running a $6.9 million health system with no regulatory framework built for that reality. HIPAA was written for insurers and providers. It has almost nothing to say about what happens when the entity holding your claims data also controls your paycheck and your performance review. The question I keep returning to: if employers are now the dominant clinical decision-makers in American healthcare, controlling roughly $800 billion in annual spending, what does informed consent even mean in that context? When your employer designs the protocol, selects the network, and receives aggregate health reports on your workforce, how much of your care is actually yours to decide? onhealthcare.tech/p/the-corporat…

English

0

4

Arianna Huffington@ariannahuff·1d

What’s the annual cost of health care for a family of four covered in a typical employer plan? As @caitlinnowens reports in @Axios: $37,824, according to a new estimate by @millimaninsight. “For as long as I've been covering health care, there's been general recognition that if any one group can really put a dent in health care spending, it's employers,” writes Owens. “And yet, employers and workers keep shelling out more for their health care.” It’s true, but the increasing alarm by self-insured employers about rising health care costs is just one of the forces converging that have the potential to finally bring change to the health care system — and not just in costs but in outcomes, too. There are consumers wanting to be more empowered to take charge of their health. There is new scientific data being published every day about the profound impact that our daily behaviors have on health outcomes – both in preventing diseases and in optimizing the treatment of diseases. And what will be transformative is when all of these forces come together with AI and its power of hyper-personalization. In recent years, the science has been increasingly clear about how our daily behaviors — food, movement, sleep, stress management and connection – aren’t just warm and fuzzy wellness practices but are among the most powerful medical interventions we have. But simply being told by a doctor to get more sleep or eat a Mediterranean diet doesn’t produce much change. Now, however, an AI health coach can hyper-personalize nudges and recommendations in real time, lowering the friction and making the healthier choice the easier choice. And the concern about rising costs of self-insured employers can help drive adoption. Given that over 60% of the population gets health care from their employers, and we spend one-third of our lives at work, the workplace can be a significant transmission mechanism not just to lower costs but to impact long-term population health. You can read Owens’ piece here: axios.com/newsletters/ax…

English

2

1

4

2.8K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The benchmark critique here is exactly right, and it connects to something worth pushing on further. Most of the field is still celebrating MedQA accuracy scores as if correctly labeling a completed clinical picture tells us anything about whether a model can function inside a live diagnostic workup, it doesn't. The sequential belief updating problem is the actual hard part, and a model that gets the final answer right through pattern matching on surface features is not doing the same cognitive work as a clinician who updates probability estimates on PE versus ACS as each new data point lands. Where this gets harder than the preprint framing suggests: even if you design a benchmark that tests sequential updating, you still have to decide what the ground truth probability distribution looks like at each step. Clinicians disagree on that, sometimes substantially, and those disagreements are not random noise, they reflect different prior weights shaped by training, patient population, and practice context. Building an evaluation framework that captures appropriate belief revision requires resolving whose beliefs count as appropriately calibrated, which is a harder epistemological problem than it sounds. That gap is what I've been arguing separates documentation AI from reasoning AI at an architectural level. LLMs are next-token predictors optimized for compression, the sequential inference problem requires persistent state representation and calibrated uncertainty quantification, and current benchmarks are nearly blind to that distinction. onhealthcare.tech/p/clinical-rea…

English

0

10

Nikita Mehandru@nikita_mehandru·19h

🩺Medical benchmarks measure if LLMs get the correct final diagnosis. True clinical reasoning requires sequential belief updating: does the model revise its beliefs appropriately as new evidence appears? New preprint: arxiv.org/abs/2505.22919

English

4

8

24

2.7K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

Burnout-driven turnover in primary care is going to hit differently over the next decade than it does today, because the physicians walking out the door are increasingly the ones who haven't yet moved into risk-bearing arrangements. Top-quartile PCPs in capitated models through platforms like Agilon and Oak Street are already pulling $700K-$900K, and that comp shift is one of the few structural forces that can actually change the retention calculus at the practice level. The problem is that the physicians most exposed to Medicare PFS compression, which dropped from $33.29 to $32.35 per conversion unit in 2025 alone and is cumulatively down roughly 30% in real dollars since 2020, are the same fee-for-service PCPs most likely to show up in burnout surveys. I looked at how these payment dynamics intersect with workforce supply in onhealthcare.tech/p/how-ai-value… and the through-line is uncomfortable: the AAMC projects an 86,000-physician shortage by 2036, with 40,000 of those in primary care, and burnout-driven exits are pressure on top of a supply gap that's already structural. Turnover data and payment compression aren't separate problems.

English

0

11

JAMA Internal Medicine@JAMAInternalMed·1d

National analysis linking US board-certified family physician survey responses with Medicare data found that self-reported #burnout was associated with significantly higher rates of physician turnover, including both changing practices & leaving practice. ja.ma/4uyFqZQ

English

1

2

8

1.1K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The recovered funds argument sounds clean but runs into a structural problem: Medicaid's improper payment rate has exceeded 20% in some years on $700+ billion in spending, which means you're not talking about a discrete pool of bad actors you prosecute once and recover. You're talking about a payment architecture with almost no prospective controls, so fraud regenerates faster than prosecution can clear it. And that's the part the "prosecute and recover" framing consistently skips. The pay-and-chase model, which is essentially what Medicare fee-for-service and Medicaid both run on, means CMS pays first and investigates later, often years later, often recovering pennies on the dollar from defendants who have already dissipated assets or dissolved shell entities. The DOJ Strike Force pulled $2.5 billion in takedowns in June 2023 alone, which sounds large until you recognize it represents a fraction of a single year's losses in a system that loses that much continuously. The more uncomfortable question is why commercial plans lose 1-3% to fraud while government programs lose 8-20%, and the answer points directly at prospective utilization controls (prior authorization, claims editing, network credentialing) that commercial plans run and government programs largely don't. Prosecution recovers some of what's already gone. Structural controls prevent the loss in the first place. Funding coverage expansions from fraud recovery assumes you can sustainably close that gap through enforcement alone, but the evidence from decades of Strike Force operations suggests you cannot without also fixing the payment architecture that makes the fraud so easy to commit. I went through the full mechanism here: onhealthcare.tech/p/prior-auth-a…

English

0

5

Wall Street Journal Opinion@WSJopinion·1d

There’s a simple way to lower healthcare costs without requiring Americans to bear unacceptable trade-offs: prosecute Medicaid fraud and put the recovered funds back into citizens’ pockets, writes @VivekGRamaswamy on.wsj.com/4dAvKqs

English

9

6

22

1.6K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The 1-in-5 number does seem high depending on how the sample was constructed (online panels skew toward engaged health consumers). But zooming out: if tirzepatide is already that penetrated, retatrutide's 28.3% mean weight loss hitting a heavier baseline population is going to blow up whatever utilization management criteria payers built around existing benchmarks. The 35% TWL responder cohort is where it gets really uncomfortable for payers, because that's surgical outcome territory and the step-therapy logic doesn't have a good answer for it yet. onhealthcare.tech/p/eli-lillys-t…

Cernovich@Cernovich

I'm a peptide maximalist, so no hate here. A lot of surprise tho. 1 in 5 are on Ozempic or Tirzepatide? I didn't realize that many people went to the doctor. Let alone took a specific category of drug.

English

0

67

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The 1-in-5 number does seem high depending on how the sample was constructed (online panels skew toward engaged health consumers). But zooming out: if tirzepatide is already that penetrated, retatrutide's 28.3% mean weight loss hitting a heavier baseline population is going to blow up whatever utilization management criteria payers built around existing benchmarks. The 35% TWL responder cohort is where it gets really uncomfortable for payers, because that's surgical outcome territory and the step-therapy logic doesn't have a good answer for it yet. onhealthcare.tech/p/eli-lillys-t…

English

0

23

Cernovich@Cernovich·22h

I'm a peptide maximalist, so no hate here. A lot of surprise tho. 1 in 5 are on Ozempic or Tirzepatide? I didn't realize that many people went to the doctor. Let alone took a specific category of drug.

English

35

9

274

37K

Thoughts on Healthcare Markets and Tech@thoughtson_tech·3h

The "behind the scenes" piece is actually where the glucagon receptor component earns its keep, and it's the part that gets the least attention in how people talk about this drug. Most GLP-1 conversation anchors to appetite suppression as the primary mechanism, which makes sense for semaglutide and even tirzepatide. But retatrutide's glucagon agonism is doing something metabolically distinct, pushing hepatic fat oxidation and energy expenditure in ways that don't necessarily translate into the felt experience of appetite control. You can be eating more or less normally and still be burning differently. The clinical signal for this is in the waist circumference reductions from TRIUMPH-1, 24.1 cm at 12 mg, which points to visceral adiposity clearance that goes well beyond what caloric restriction alone explains. That gap between subjective experience and what the data shows is probably going to matter a lot when payers try to build utilization management criteria around patient-reported outcomes. If the mechanism doesn't feel like "suppression" to patients, some UM frameworks will struggle to verify adherence or therapeutic response using the proxies they've built around the GLP-1 class. What I'm still working through is whether that dissociation between felt effect and metabolic effect creates a compliance problem or a compliance advantage in the long run... Full breakdown on the TRIUMPH-1 data and what it means for payer logic and the surgery comparison: onhealthcare.tech/p/eli-lillys-t…

English

0

10

💯 fitcap 💯@fitcapdefi·1d

Retatrutide doesnt do so much for appetite suppression to me either at this point, but what it does do is keep me on track and dialed in with zero effort besides the host of other benefits happening behind the scenes

Marlon@drmarlonperalta

Life changes when you realize this about retatrutide. Reta does zero for my appetite (even at 10mg). I used to complain about it but it revealed how powerful it is for muscle gain. I can eat ~500g carbs while insulin sensitivity and lipids improve WITHOUT the fat gain.

English

7

1

48

9.3K

Thoughts on Healthcare Markets and Tech

Keşfet