Disclosure: I am a physician, and physician payment affects me directly. Where a source has a stake in the outcome, including the AMA, the AAMC and the American Hospital Association, I have said so.
A study that promises universal coverage, 114,000 lives saved and a trillion dollars in savings every year deserves a careful reading. I have gone through the release and the assumptions the preprint reports. My conclusion is that it is a careful accounting of what a single-payer system would save, and it does not model whether the clinicians exist to deliver the care.
My larger concern is how the study frames its answer. Yale’s release quotes the authors as saying a system that covers everyone and costs less requires no new discovery, only enactment. That treats one of the hardest problems in American medicine, who will deliver the care, at what pay, trained by whom and within what limits, as a matter of passing a bill. It is a narrow accounting answer presented as a complete one. The American Action Forum described the analysis as widely publicized, at a time when candidates across the country were voicing support for exploring Medicare for All. In a headline, a preprint with stated limitations becomes evidence that a goal is affordable and achievable. I think that is a dangerous way to present it: it can be used to substantiate goals whose feasibility the model never tested.
That is not a small omission. Coverage is a payment arrangement. Care is a doctor, a nurse and a bed, and the model does not address where more of them would come from, or what happens to the ones already here when their payment falls.
What the Yale study claims, and how the headline is built
In August, the Yale School of Public Health announced a preprint projecting that a single-payer system could cover every American, avert about 114,000 deaths a year, and cost roughly $1 trillion less than the system it replaces. The preprint has not been peer reviewed. The authors describe the results as a lower bound in several respects.
I have spent more than thirty years in gastroenterology, and I approach a study like this the way I approach any model: by looking at what it counts, and then at what it leaves out.
What it counts. The model is a static, one-year accounting exercise built on 2024 national health spending of about $5.3 trillion. Its gross reductions come from five lines:
Source of savingsApprox. annual amountDrug prices cut 51% (international reference pricing)$378 billionAdministrative overhead falls toward Medicare’s level$286 billionFraud reduced 8%, extrapolated from Taiwan’s single-payer transition$286 billionAll hospital and clinical fees paid at Medicare rates$296 billionAvoided emergency and inpatient care$100 billion
Against roughly $1.35 trillion in gross reductions, the model adds back about $304 billion for care that uninsured and underinsured people would use, care that now goes unpaid, and universal dental coverage. The net is about $1.04 trillion. According to Yale’s release, under more conservative assumptions about drug prices and fraud reduction, the floor is about $663 billion.
Two things stand out. First, the drug price cut is the largest single line, and together with the fraud reduction it accounts for about half of the gross reductions. Both depend on policy and legal outcomes and on an extrapolation from another country, not on anything the model measures. The provider payment cut is the second-largest line, at roughly $296 billion. Second, the savings are computed on a health system that is assumed to keep functioning exactly as it does now.
How the 114,000 lives are built. The headline combines two different things:
About 62,900 deaths averted come from universal coverage itself. Of these, roughly 33,000 are among the uninsured and roughly 29,600 among the underinsured. For the underinsured there is no direct estimate of excess mortality, so the authors interpolated a hazard ratio of 1.25, within a plausible range of 1.0 to 1.4, and applied it to adults aged 19 to 64.
About 51,300 more come from reversing coverage rollbacks and other policies enacted since 2025. That figure is taken from a prior projection and is not an effect of single-payer.
So roughly 45% of the headline number would be achieved by undoing recent policy, with or without Medicare for All. A fair summary is that the coverage-driven benefit is real but smaller and more uncertain than “114,000 lives” suggests.
The authors’ own caveats. To their credit, the authors say what the model leaves out: transition costs, administrative job losses, and how providers would respond to Medicare payment rates. Those are the pieces that decide whether the plan works in practice. They are also nowhere in the headline.
The payment cut is bigger for some than others
The model assumes every hospital and clinician is paid at Medicare rates. That is a very different proposition for a hospital than for a physician practice, and the two should not be blurred together.
Hospitals. RAND’s 2022 analysis found commercial hospital prices averaging about 254% of Medicare, a figure the American Hospital Association disputes on methodology. KFF’s literature review put hospital services at roughly 199%. On either number, a move to Medicare rates is a very large cut in revenue per service.
Physicians. The gap is smaller. KFF’s review of the literature put physician services at about 143% of Medicare, with a range of 118% to 179%. For gastroenterology, one industry-consultant estimate is about 120%, though KFF notes that procedure-heavy specialties tend to run higher. Moving the commercial portion of a physician’s revenue to Medicare rates would cut that portion by roughly 30% overall and about 17% in GI, by my arithmetic from those ratios. That is substantial, but it is not the 2.5-times figure that gets quoted for hospitals. The cut is not uniform, either: Medicaid generally pays physicians less than Medicare, so practices with many Medicaid patients could gain, and the net effect depends on each practice’s payer mix.
Medicare itself is already a moving target. The physician fee schedule conversion factor stood at $36.09 in 2020 and fell to $32.74 by 2024, a nominal drop of about 9%. For 2026 it rose to about $33.40, but that includes a one-year 2.5% increase that expires. The 2027 rule, as proposed in July 2026, would lower it by about 1.7% for most physicians, because the increase lapses. I have not confirmed whether the final rule has been released. The AMA says Medicare physician pay has fallen 33% since 2001 when adjusted by the Medicare Economic Index, a measure of practice costs.
I want to be fair about that last figure. The AMA is an interested party, the comparison measures pay per unit of work and not total income, and MedPAC found that median physician compensation rose in 2022 and that Medicare beneficiaries’ access was comparable to that of the privately insured. The honest version of the argument is narrower and, I think, stronger: the baseline is already eroding on a published schedule, and the Yale model proposes a further step down without asking how practices would absorb it.
What happens to the practices. Independent practice is already shrinking. In AMA surveys, physician-owned practices fell from 60% of physicians in 2012 to 42% in 2024, while hospital-owned practices rose from 23% to 35%. Private equity ownership is about 6.5%. In AMA surveys, physicians cite insurer payment and regulation as reasons for leaving private practice. Whether lower rates would accelerate that shift is a question the model does not ask, and consolidation has its own price effects. One 2025 study found colonoscopy prices about 4.5% higher after private equity acquisition of GI practices.
The evidence on closures after a rate cut is thinner than advocacy surveys suggest, and I won’t overstate it. Medicare opt-out remains low, about 1.2% of physicians in 2024. After the Balanced Budget Act of 1997, MedPAC found no meaningful physician-access problem, though hospitals and home health agencies were hit hard. The honest reading is that physicians have absorbed past cuts. But the comparison cuts both ways: many scheduled cuts in the SGR era were overridden by Congress, so they were threatened more than realized, and they are weak evidence of what a permanent reset would do.
What the model assumes happens to the money. For hospitals specifically, the American Hospital Association says Medicare paid about 83 cents per dollar of cost in 2023. Jessica Riedl of Brookings called the Yale work an accounting exercise that did not model reality, and Michael Baker of the center-right American Action Forum argued that hospitals would specialize or cut services. Both are describing the limitation the authors themselves acknowledge: roughly $300 billion less provider revenue, with no modeled response.
Who delivers the care?
Cutting what the system pays is one question. Whether the system has the people to meet higher demand is another. The Yale model prices demand for the newly insured and underinsured, at the rate insured patients now use care, and does not ask who would provide it. The model’s own add-back for that new use is about $198 billion, under 4% of national spending. That sounds small, but it would land on a physician workforce that HRSA projects will meet only about 88% of demand by 2038, before any expansion.
Physicians. The AAMC, which advocates for more federally funded residency positions, projects a shortage of between about 13,500 and 86,000 physicians by 2036, and that is without a coverage expansion adding demand. HRSA’s projection is larger: about 141,000 full-time-equivalent physicians short by 2038, with supply meeting about 88% of demand. The workforce is aging. In 2021, about 17% of physicians were 65 or older and another 25% were 55 to 64, and more than a third are expected to retire within a decade.
The pipeline is slow. Medicare-funded residency positions have been capped since the Balanced Budget Act of 1997, apart from small increases added in recent years. Medical school enrollment passed 100,000 for the first time in 2025, and the 2026 Match had a record 44,344 positions, so the bottleneck is not applicants. But four years of medical school plus three to seven years of residency, with fellowship on top for some, means new capacity takes most of a decade or more to arrive. Whatever the coverage law does, the supply that has to meet it is already largely set.
Nurses. HRSA projects about a 10% shortage of registered nurses from 2027 through 2037 (24% in non-metro areas in 2027) and licensed practical nurse supply at only 64% of demand by 2037. Nursing schools turned away 92,672 qualified applications in 2025, citing faculty, clinical sites and preceptors. Bedside RN turnover was about 17.6%.
Imported capacity is under strain. International medical graduates are more than a quarter of US physicians and about 30% of residents. Policy has been moving the wrong way: a 2025 pause in J-1 interviews, expanded travel restrictions, and a $100,000 fee on new H-1B petitions that, according to press reports, a federal judge struck down for physicians in June 2026, with the scope unclear and an appeal possible. In 2026, visa-requiring non-citizen IMGs matched at 54.4%, a five-year low.
What happened in earlier expansions. The evidence is imperfect, but it suggests longer waits. In Massachusetts after 2006, a physician-society survey in 2011 found about half of internists and family physicians not accepting new patients and internist waits of 48 days. That is a descriptive survey, not a controlled study. After the ACA’s Medicaid expansion, a ten-state secret-shopper study found Medicaid appointment availability rose, while the share of privately insured patients waiting 30 or more days rose from 7.1% to 10.5%. Supply responded mostly through longer waits and redirected capacity, not quick new supply.
Retirement. In one claims study, the annual rate at which physicians stopped billing Traditional Medicare doubled to 3.6% between 2010 and 2024. That measure also captures moves to Medicare Advantage, not only retirement, and the authors do not attribute it mainly to payment cuts. I found no rigorous study showing that cuts cause retirement. It is plausible, and unproven, that a payment cut pulls retirements forward among the 42% of physicians who are 55 or older. That is exactly the question the model should have asked.
What peer countries do differently
The usual reply is that other wealthy countries pay doctors less and still deliver care. That is true, and it is worth being precise about what comes with it.
Pay. Medscape’s 2023 international survey, which is self-reported and converted at market exchange rates, put average annual physician pay at about $352,000 in the US, $273,000 in Canada, $160,000 in Germany, $122,000 in the UK and $93,000 in France. The gap is mostly a specialist gap. In Medscape’s 2021 edition, US general practitioners averaged about $242,000 against about $200,000 for German GPs.
Training. In peer systems, the state pays for much more of the path to becoming a doctor:
Germany’s public medical schools are essentially tuition-free, with small semester fees, and residents are salaried.
In England, tuition is capped at £9,250 a year, graduates can owe up to about £100,000 with maintenance loans, and the loans are income-contingent. Trainees are salaried by the NHS.
In the US, AAMC data as reported by aggregators show a median education debt of about $215,000 among graduates with debt, including premedical borrowing (about $200,000 for medical school alone), with 70% of graduates borrowing and 28% owing $300,000 or more. The four-year median cost of attendance for the class of 2026 is about $298,000 at public schools and $408,000 at private schools. Residency is salaried, and Medicare supports it with about $21 billion a year, but medical school itself is paid for privately.
My point is not that American doctors are entitled to their current pay. It is that the compensation and the cost of training are linked, and the Yale model changes one without the other. A system that pays physicians at peer-country levels while asking them to finance their education at US prices is a different bargain from the one those systems offer. Someone pays for a physician’s training: the public up front, as in Germany and England, or the physician later through pay, as in the US. A model that cuts the pay but never says who pays up front has not removed the cost, only left it off the page.
Emigration. I could not find reliable counts of how many peer-country doctors come to the US and stay, or how many return, and I won’t invent them. What the data show is more mixed than the “brain drain” line suggests. Ireland’s statistics office found about 30% of one graduate cohort left the country at some point, and only 39% of those were back after ten years. In the UK, about 42% of doctors qualified abroad, and 4,880 non-UK-qualified doctors left UK practice in 2024. Canadian physician outflow to the US has been shrinking since the 1990s. The larger structural fact is that peer systems depend on imported doctors too.
Strain in peer systems. Single-payer and universal systems have their own capacity problems. Canada’s median wait from GP referral to treatment was 28.6 weeks in 2025, according to the Fraser Institute, a market-oriented think tank that relies on physician self-reports. England’s NHS waiting list was 7.33 million in July 2026, with 65.3% treated within 18 weeks. England’s resident doctors went through 15 rounds of strikes before accepting a deal in June 2026 that includes 4,500 additional training posts. These systems show that coverage does not guarantee timely access when capacity is tight.
The fair counterpoint. The Commonwealth Fund’s Mirror Mirror 2024 ranked the US last of ten countries overall, last on health outcomes and administrative efficiency, and second on care process, while spending more than 16% of GDP against roughly 8 to 12% elsewhere. Peer systems get comparable or better results for less. My own view is that the US would need to expand training and capacity alongside any payment change, not after it.
The drug bill
Drug spending raises the same capacity question. Medicare Part D gross spending on GLP-1 drugs, before rebates and mostly for drugs labeled for diabetes, rose from $7.1 billion in 2020 to $27.6 billion in 2024, and Medicaid spending from $1.5 billion to $8.8 billion. Gallup found 12.4% of adults taking a GLP-1 for weight loss in 2025. The Yale model prices drugs on a static 2024 base. Reference pricing would cut what these drugs cost per prescription, but with no cost sharing their use would likely keep growing, and a one-year model cannot show which effect wins. Obesity drives much of this demand, and its roots in the food supply deserve their own analysis, which I will publish separately. The short version is that the national data point to how much and what Americans eat, not to any single ingredient such as high fructose corn syrup.
Is the drug worth it, and what would we pay?
A note on cost growth. Medicare’s physician conversion factor rose less than 8% in nominal terms from 1992 to 2026 ($31.00 to $33.40, including the expiring one-year increase), while total spending kept climbing. I would argue the physician fee schedule has not been what drove spending up. Total spending also depends on volume and on commercial prices, and the Yale model itself counts the move to Medicare rates for all providers as its second-largest saving, at about $296 billion.
What the drug evidence supports. ICER’s December 2025 review found semaglutide and tirzepatide cost-effective at $100,000 per quality-adjusted life-year (QALY) at net prices of about $6,800 and $8,000 a year. An independent lifetime model in JAMA Health Forum put tirzepatide at $197,000 and semaglutide at $468,000 per QALY at net price. An American College of Physicians review of non-industry studies rated tirzepatide high value in moderate-certainty studies and liraglutide low value, and found no high-certainty comparison. So the answer depends on price and on how long people stay on treatment. Across claims and health-system studies, between about 19% and 56% of patients were still on a GLP-1 at one year, and in one trial, patients who switched from tirzepatide to placebo regained much of the weight. CBO estimated about $35 billion in net federal cost for 2026-2034 if Medicare covered these drugs. The best evidence that the drug reduces other care comes from SELECT, in people with established heart disease: about 10% fewer admissions. I found no study showing full cost offsets in people with obesity but no cardiovascular disease.
What would we pay? The US government already has several numbers for a life. The Department of Transportation’s value of a statistical life is $14.2 million (2025 base year). HHS uses about $14.1 million, which converts to roughly $611,000 per life-year and $726,000 to $1.2 million per QALY. ICER’s benchmark is $100,000 to $150,000 per QALY, and the UK’s NICE uses £25,000 to £35,000. Empirical estimates of a US threshold vary and use different units: about $104,000 per QALY in an opportunity-cost study (Vanness) and $183,000 to $264,000 per life-year as an implied US decision rule (Braithwaite). These figures measure different things. The regulatory values reflect what people reveal they would pay to reduce their own risk, while health-care thresholds such as ICER’s and Vanness’s estimate what the same money would buy elsewhere in the system. Adopting a transportation-style value in health care would raise what the system pays, not lower it: at $726,000 or more per QALY, every GLP-1 estimate above would count as cost-effective.
The law restricts the lower figure. Federal statute bars the research institute created by the Affordable Care Act from using a cost-per-QALY threshold, and the Inflation Reduction Act bars treating life extension for elderly, disabled or terminally ill people as lower value. Critics of QALY thresholds argue that they disadvantage older and disabled patients. England uses one anyway, through NICE, as an open way to set limits. I take no position here on whether the US should.
Non-drug options. Modelling studies generally find sugar-sweetened beverage taxes cost-saving or about $20,000 to $43,000 per QALY, though California’s is about $125,000 with a wide interval. Calorie labeling is modelled as cost-saving and the UK levy as a net benefit, and bariatric surgery is generally cost-effective. Nearly all of this is simulation, not trial data.
Why this matters for the Yale model. A system that removes cost sharing and covers everyone has to say what it will and will not pay for, at what price per year of life, and who decides. The preprint does not address it. England answers it explicitly through NICE. In my experience, the US makes the same choices less visibly, through coverage rules, prior authorization and what patients pay. That question belongs in the analysis before the savings are announced.
What a serious analysis would ask
None of this means universal coverage is unachievable or that the current system is working. Millions of uninsured Americans, and the more than 45 million working-age adults the study counts as underinsured, are real problems, and the authors make a serious argument that the country already spends enough and the issue is how the money is allocated. The question is what a responsible model of that change would include. At a minimum:
A supply constraint. Model how many visits, procedures and bed-days the existing workforce can deliver, and report access and wait times as outputs next to cost.
Provider response to the payment schedule. Estimate retirements, practice closures, hospital service cuts and consolidation at the assumed rates, with a range, instead of assuming care continues unchanged.
Separate the coverage effect from other policy. Report the deaths averted by coverage apart from those attributed to reversing unrelated 2025 policies.
Treat drug and fraud savings as scenarios. A 51% drug price cut and a Taiwan-derived fraud reduction are policy and legal outcomes, and together they are about half of the gross savings.
A transition path. Include the cost and time of moving people off existing coverage and the administrative job losses the authors acknowledge but do not model.
A training-pipeline plan. If universal coverage is the goal, the model should say what funds the residency slots, nursing faculty and clinical sites that the new demand will require.
The preprint does acknowledge several of these as limitations. My criticism is that the release leads with the savings and the lives and places the limitations lower down.
The strongest case on the other side
A critical piece should say what a fair opponent would say.
Supply limits cost, as well as access. Gaffney and colleagues, who are affiliated with the advocacy group PNHP, reviewed coverage expansions in the US and ten other countries. They report that use rose far less than most models predicted, ambulatory visits by about 7 to 10%, and that supply limits push providers to triage toward higher-need care. CBO’s 2020 analysis found national health spending under single-payer options ranging from $0.7 trillion lower to $0.3 trillion higher in 2030. On this view, the shortage argument overstates the surge.
Rates are a choice. Proponents note that legislation could set a fee schedule above current Medicare rates, and PNHP argues that billing reductions could free about 5% of physician hours. A summary of a Mercatus analysis of Sanders’s plan estimated blended physician pay would fall about 5 to 10%, and about zero net if volume rose 11%.
The status quo has its own access problem. MedPAC’s 2025 survey found 29% of Medicare patients waited six weeks or more for a new primary care visit and 31% for a new specialist visit. Coverage that is not affordable is not access either.
The authors are explicit. They state the limitations and frame the results as a lower bound in several respects, including the unmodeled billing savings, and they count underinsurance deaths only up to age 64 even though most such deaths occur later.
I think those arguments are serious. They also cut both ways: if supply limits keep cost down by limiting how much care is delivered, then they also limit how many of the 62,900 coverage-driven deaths can actually be averted. The deaths-averted estimate assumes the newly covered receive the care. Whether the workforce can deliver it is a separate question, and the model does not answer it.
The bottom line
The release’s claim that only enactment is needed gets the problem backwards. Passing a law is the start, not the solution. What the model does not show is the supply: the physicians retiring over the next decade, a training path of roughly a decade from medical school to independent practice, the nurses and nursing faculty already in short supply, and the international graduates whose path into the country is narrowing.
That is why the presentation matters. A model that shows savings and lives saved, without showing capacity, gives policymakers and the public a reason to set goals the delivery system cannot meet. A model this consequential should have been built around the delivery system, not just the budget. The cost of that mistake would fall on patients waiting for care that coverage promised and the workforce could not provide.
In my judgment, cutting what the system pays while adding demand is likely to worsen the shortage. That is an argument about consequences, not a modeled result.
The honest way to proceed is the reverse of the headline. Build the training pipeline, model the supply response, separate the coverage effect from other policy, and then ask what the system can afford to pay the people who deliver the care, and what it will pay for at all, at what price per year of life.
Sources are listed below. Figures are from the cited pages and agency reports; where a source is an advocacy group or interested party, I have said so in the text.
Sources
The Yale study and its critics
Payment and practice economics
AMA on Medicare physician pay since 2001 (interested party)
Workforce
International comparisons and training costs
Drugs and value of life