Health Policy & Management

What 4.9 trillion dollars buys

In 2023 the United States spent about 4.9 trillion dollars on health care. That works out to roughly 14,570 dollars for every person in the country, and about 17.6 percent of everything the economy produced that year. Germany, whose system covers everybody, spent a little under 12 percent of its output. The United Kingdom spent less still. In the same year American life expectancy at birth was 78.4 years, which put it near the bottom of the wealthy countries.

Hold those numbers next to each other and you have the question this course exists to answer. It is not why American health care is expensive, which is a question with an answer you can look up. It is a harder one: what is a health system actually for, what jobs does it have to do, and how would you know whether a given arrangement is doing them?

You cannot answer that by starting with the United States. American health care is an unusually complicated instance of a general thing, and if you learn the instance first you will mistake its accidents for necessities. So this lesson starts one level up. Every country that delivers medical care to a population, whether it does so through a national service, a set of nonprofit sickness funds, or a patchwork of employer plans and government programs, has to solve the same short list of problems. Learn the list, and the rest of the course becomes a set of variations on it.

The core of it: Health systems differ enormously in their structures and hardly at all in their tasks, so the productive comparison is not between institutions but between the ways different countries perform the same handful of functions.

Five questions every system answers

Here is the shortest useful version of the list. Every system, including one nobody designed, gives an answer to each of these. Sometimes the answer is deliberate policy; sometimes it is whatever fell out of history. It is still an answer.

Question	What it decides	Where you look to find the answer
Who is covered?	Whether coverage is universal, categorical, or earned through employment	Eligibility rules, enrollment statutes, uninsured rates
What is covered?	The benefit package: which services, drugs, and devices are in and which are out	Benefit schedules, formularies, coverage determinations
Where does the money come from?	Taxes, payroll contributions, premiums, or payment at the point of service	National health accounts, budget documents
How are providers paid?	Whether the incentive is to do more, to do less, or to do something else	Fee schedules, contracts, global budgets
Who decides all of the above?	The locus of authority: a ministry, a legislature, a regulator, a market, or several at once	Enabling legislation, regulatory jurisdiction, court decisions

Try the list on a system you already half know. In the United Kingdom the answers are: everyone ordinarily resident, a comprehensive package defined largely by the National Institute for Health and Care Excellence and local commissioners, general taxation, a mixture of salary and contract with global budgets, and a health department answerable to Parliament. In the United States the answers are: it depends on your age, income, employer, state, and immigration status; it depends on your plan; it comes from employers, payroll taxes, general revenue, premiums, and your own pocket; it depends on the payer and the service; and authority is split among Congress, dozens of federal agencies, fifty state insurance departments, courts, accreditors, and private contracts.

Notice what that second answer tells you before you have learned a single American acronym. A system with five answers to every question will spend a great deal of money on figuring out which answer applies to which person, and it will leave people in the gaps between answers. Both of those predictions turn out to be true, and Module 2 will put numbers on them.

Why this matters: The five questions are a diagnostic tool. Ask them of any arrangement and you will locate its coverage gaps and its administrative burden before you know anything about its history.

The WHO functions, and why they are more useful than they look

The World Health Organization's framework, developed in its 2000 report on health system performance and refined in 2007 into what practitioners call the six building blocks, breaks the same territory into a different set of parts: service delivery, the health workforce, health information systems, access to essential medicines and technologies, financing, and leadership and governance.

The building blocks look like a bureaucratic list and are frequently taught as one, which wastes them. Their value is that they force you to notice the parts of a system that never appear in political argument. Health financing is argued about constantly. Health information systems are argued about almost never, and yet a country that cannot tell you how many people have diabetes cannot plan for diabetes, and a hospital that cannot retrieve last month's chart will repeat last month's imaging. The workforce is similar: you can legislate coverage for forty million people in an afternoon and you cannot train a nurse in less than two years or a specialist physician in less than a decade.

The 2000 report also proposed three goals against which the functions should be judged: better health, responsiveness to what people legitimately expect from a health system apart from health itself, and fairness in financial contribution, meaning that households should not be ruined by illness. Those three are worth memorizing because they explain why disagreements about health policy so often fail to resolve. Two people who agree completely about the facts can disagree about whether a system is good because they are weighting the three goals differently.

Remember: The functions include the invisible ones. Information systems, supply chains, and workforce pipelines determine what a policy can actually deliver, and they move on timescales of years, not legislative sessions.

Why medical care is not a normal market

In 1963 the economist Kenneth Arrow published a paper in the American Economic Review called Uncertainty and the Welfare Economics of Medical Care. It is the founding document of health economics, and its argument is worth having in full rather than in slogan form, because both the pro-market and the anti-market camps quote it and both leave out the half they dislike.

Arrow's method was not to declare medical care special. It was to take the conditions under which a competitive market produces an efficient outcome and check them one by one against medical care. Several fail. Demand is irregular and unpredictable, so an individual cannot plan for it the way they plan for food or housing. The product is not a good you can inspect before buying; you are buying a judgment about what you need, from the person who will then supply it. Information is deeply asymmetric, and the buyer often cannot evaluate the product even after consuming it, because you generally cannot tell whether you recovered because of the treatment or despite it. Entry is restricted by licensure. And crucially, the risk itself cannot be traded on an open market without the insurance problems you will meet in the next lesson.

Arrow's conclusion was subtle. The nonmarket institutions that surround medicine, professional licensure, ethical codes that forbid advertising outcomes, trust as an explicit substitute for verification, and nonprofit ownership of hospitals, are not irrational impositions on a market. They are the institutional response to those specific market failures. Wherever the market cannot function, some other mechanism grows to do the job, and you should expect to find one.

What Arrow did not say is that markets have no role in medical care. Prices allocate; competition can discipline. His point is that the standard proof that a market gets it right does not go through here, so every claim that competition will fix a health care problem has to be argued from evidence about that specific problem rather than assumed from theory. You will apply that test repeatedly, most sharply in the lessons on payment models and drug pricing.

Key idea: Arrow showed that the specific assumptions behind efficient markets fail in medical care, which is why medicine is wrapped in licensure, ethics, trust, and nonprofit ownership, and why market fixes must be argued case by case rather than assumed.

Kissick's triangle, and the trade you cannot avoid

The physician and health policy scholar William Kissick, writing in 1994, described what he called an iron triangle with three corners: access, quality, and cost containment. His claim was that you can improve any two only at the expense of the third. Expand access and hold quality constant, and spending rises. Hold spending constant and expand access, and something about quality or the breadth of what is covered has to give.

Treat the triangle as a discipline rather than a law of nature. It is not literally impossible to improve all three at once; eliminating a genuinely useless test improves cost without touching access or quality, which is why waste reduction is the one policy everybody claims to support. But genuinely wasteful care is harder to identify prospectively than it looks, and one person's waste is another person's reassurance. The triangle's real function is to force a question at the end of every policy proposal you read: which corner is this proposal quietly giving up, and does it say so?

Work a case. A state proposes to cover all its uninsured adults under Medicaid without new revenue. Access improves sharply. The cost corner has to absorb it somewhere: through lower payment rates to physicians, which is a quality and access problem in disguise if physicians stop taking Medicaid patients; through narrower benefits; through cuts elsewhere in the state budget; or through a new tax that the proposal did not mention. There is no version in which nothing gives. A proposal that claims otherwise has hidden the transfer, and finding it is your job as a reader.

The upshot: Every serious health policy proposal moves at least one of access, quality, and cost in the wrong direction, and the honest ones say which one and why they consider the trade worth making.

Three systems against the same checklist

Here is what the framework buys you. Three countries, wildly different institutions, compared on the same five questions.

	United Kingdom	Germany	United States
Who is covered	All ordinary residents, automatically	Nearly everyone, through statutory sickness funds, with a private tier	Categorically: by age, income, employment, veteran status, state
Money comes from	General taxation	Payroll contributions shared by worker and employer, plus federal subsidy	Employers, payroll taxes, general revenue, premiums, out of pocket
Providers paid by	Salary, capitation for general practice, national tariffs	Negotiated fee schedules and hospital case rates	Fee schedules, negotiated commercial rates, capitation, bundles
Who decides	Parliament and the health department, with national appraisal bodies	Corporatist self-government: funds and physician associations negotiate within a legal frame	Congress, agencies, fifty states, courts, employers, private contracts
Characteristic complaint	Waiting times for elective care	Complexity of the two-tier structure and rising contribution rates	Cost, coverage gaps, and administrative burden

Read the last row, because it is the payoff. Each system's characteristic complaint is the predictable consequence of its own design. A tax-funded system with a fixed budget rations by queue, so its complaint is waiting. A system built on many payers competing for the same providers has no single price and no single rulebook, so its complaint is cost and complexity. Nobody chose those complaints, and nobody can simply remove them without changing the structure that produces them. Module 5 works this comparison properly, with the country cases in detail.

Common misconceptions

• Some countries have a health system and others just have chaos. An unplanned arrangement is still a system in the sense that matters here: it performs the functions and produces predictable results. Calling it chaos hides the fact that its outcomes are the outputs of identifiable rules.
• Health policy is mainly about insurance. Financing is one function among several. A country with perfect coverage and no clinicians has no health care, which is why the workforce lessons in Module 3 matter as much as the financing ones.
• More spending means better health. Across wealthy countries the correlation flattens out sharply once basic access is achieved. The United States is the standing demonstration that additional spending can buy prices rather than health.
• Arrow proved that markets do not work in health care. He proved that the standard argument for markets does not go through, which is a narrower and more useful claim. It shifts the burden to evidence rather than closing the question.
• Waste reduction escapes the iron triangle. Some does. But waste is far easier to identify in retrospect than to remove prospectively without also removing care that someone needed, which is why decades of waste campaigns have not bent the spending curve much.

Where this leaves us

• The United States spent about 4.9 trillion dollars on health care in 2023, roughly 14,570 dollars per person and about 17.6 percent of GDP, while its life expectancy at birth was 78.4 years.
• Every system answers five questions: who is covered, what is covered, where the money comes from, how providers are paid, and who decides.
• The WHO building blocks add the functions that political argument ignores: workforce, information systems, and supply of medicines and technologies.
• The 2000 WHO report's three goals are health, responsiveness, and fairness in financial contribution, and people who weight them differently will disagree about a system even when they agree about the facts.
• Arrow's 1963 paper showed which competitive-market assumptions fail in medical care and explained the nonmarket institutions that grew up in their place.
• Kissick's triangle of access, quality, and cost is a reader's discipline: find the corner a proposal is giving up.

Sources

1. Centers for Medicare and Medicaid Services. (n.d.). National Health Expenditure Data. cms.gov
2. KFF. (n.d.). Health Costs. kff.org
3. Organisation for Economic Co-operation and Development. (n.d.). Health. oecd.org
4. Arrow, K. J. (1963). Uncertainty and the welfare economics of medical care. American Economic Review, 53(5), 941-973.
5. Kissick, W. L. (1994). Medicine's Dilemmas: Infinite Needs Versus Finite Resources. Yale University Press.

How Harvard broke its own insurance market

In 1995 Harvard University changed one line in how it subsidized employee health coverage. Previously it had paid a larger dollar amount toward the more generous plan, a preferred provider organization with wide choice, than toward the cheaper health maintenance organizations. The new policy paid the same dollar amount toward every plan, so employees who wanted the generous plan paid the full difference out of their own pockets.

What happened next is the cleanest natural experiment in health insurance ever to occur inside a single payroll office. Younger and healthier employees, who valued wide provider choice least and noticed the new price most, left the PPO for the HMOs. The people who stayed were, on average, sicker. Their claims were higher. The next year's PPO premium rose to cover them, which pushed out the next healthiest tier, which raised the premium again. Within three years the PPO was gone. The economists David Cutler and Sarah Reber documented the sequence in 1998 and gave the mechanism its clearest empirical demonstration.

Nobody behaved badly. Harvard was trying to make employees cost-conscious, which is a respectable goal. Employees were choosing the plan that suited them, which is what choice is for. And the market still collapsed, because insurance markets can fail in a way that grocery markets cannot. This lesson explains why, and it is the most portable lesson in the course: everything in Modules 2 and 5 is an application of it.

The point: An insurance market can destroy itself through the ordinary self-interested choices of well-informed participants, which is why insurance is regulated everywhere and left alone nowhere.

What insurance actually does

Insurance is not a way of paying for health care. It is a way of converting an uncertain large loss into a certain small one. Keep that distinction in front of you, because most bad arguments about health insurance come from forgetting it.

Work the arithmetic. Take 10,000 people. In a given year, suppose 9,000 of them incur no medical costs, 900 incur 5,000 dollars, 90 incur 50,000 dollars, and 10 incur 400,000 dollars. Total spending is 4.5 million plus 4.5 million plus 4 million, which is 13 million dollars. Divide by 10,000 people and the average is 1,300 dollars. That average is the actuarially fair premium: the price at which the insurer collects exactly what it expects to pay out.

Now look at the distribution rather than the average, because the distribution is the whole point. Ninety percent of these people spend nothing. Ten people, one in a thousand, account for 4 million dollars, nearly a third of the total. That extreme skew is a permanent feature of medical spending in every country that measures it, and it produces two consequences you should carry through the whole course. First, insurance is valuable precisely because you cannot know in advance whether you are one of the ten. Second, any policy that focuses on the average patient is aiming at almost nobody.

Why would a person pay 1,300 dollars to avoid a gamble whose expected cost is also 1,300 dollars? Because losses hurt more than equivalent gains help. A person who is risk averse prefers a certain 1,300 dollar loss to a gamble with the same expected value, and will pay something extra, called a risk premium, to make the swap. That willingness is what funds the insurer's administration, reserves, and profit, collectively the loading. Real premiums equal expected claims plus loading, which is why the medical loss ratio rules you will meet in the ACA lesson exist: they cap the loading as a share of the premium.

Worth holding on to: Insurance sells certainty, not care. Its value comes from the extreme skew of medical spending, in which a small fraction of people account for most of the cost and nobody knows in advance who they are.

Adverse selection, and the spiral it produces

Now break the model in the one place that matters. Suppose people know more about their own likely spending than the insurer does. That is not an exotic assumption; you know about your back pain, your family history, and the specialist appointment you have been putting off, and the insurer does not.

George Akerlof's 1970 paper on the market for used cars gave the general form of the problem. If sellers know which cars are defective and buyers do not, buyers will only pay the average value; owners of good cars will refuse to sell at the average; the remaining pool gets worse; the price falls again. Michael Rothschild and Joseph Stiglitz worked out the insurance version in 1976. Apply it here.

1. An insurer offers coverage at 1,300 dollars, the average expected cost.
2. People who privately expect to spend less than 1,300 dollars find that a bad deal and decline. People who expect to spend more find it a good deal and buy.
3. The enrolled pool is now sicker than the population, so actual claims per enrollee exceed 1,300 dollars. Say they come in at 2,400.
4. The insurer raises the premium to 2,400 dollars, which is now unattractive to the next healthiest layer, who drop out.
5. Repeat. In the limit only the sickest buy, at a premium equal to their own expected cost, which is not insurance at all.

That sequence is the death spiral, and Harvard ran a live version of it. Note what it is not. It is not fraud, and it is not people gaming the system in any blameworthy sense. Every actor is responding rationally to a price. The failure is structural.

Every country's insurance rules are, in large part, an answer to this problem, and there are only a few available answers. You can compel participation, so the healthy cannot leave. You can subsidize premiums enough that buying is rational even for the healthy. You can restrict when people may enroll, so nobody can wait until they are sick. You can pool by something unrelated to health, which is what employment-based coverage accidentally does. Or you can risk-adjust, paying insurers more for enrolling sicker members so that avoiding sick people stops being profitable.

Tool	How it blocks the spiral	Where you will meet it
Mandate to be insured	Removes the healthy person's exit option	ACA individual mandate through 2018; Germany, Netherlands, Switzerland
Premium subsidy	Makes buying rational at every health status	ACA premium tax credits; employer contributions
Limited open enrollment	Stops people buying only once they need care	Marketplace open enrollment; Medicare Part B late-enrollment penalty
Group pooling by employment	Sorts people by job, a trait weakly related to health	Employer-sponsored insurance, covering most working-age Americans
Risk adjustment	Pays insurers more for sicker enrollees, removing the gain from avoiding them	Medicare Advantage, ACA marketplaces, Dutch and German funds

One implication is worth stating flatly because it is the single most argued point in American health policy. Guaranteed issue, the rule that an insurer must sell to anyone, and community rating, the rule that it must charge everyone a similar price, are exactly the conditions under which adverse selection is strongest. Adopt them without a mandate or a large subsidy and you have built the spiral into law. That is not an ideological claim; it is arithmetic, and every country that has tried the combination has discovered it.

In short: Adverse selection is what happens when buyers know their own risk better than sellers do, and only compulsion, subsidy, enrollment limits, unrelated pooling, or risk adjustment can prevent the resulting spiral.

Moral hazard, and the experiment that measured it

The second insurance problem runs the other direction. Once you are insured, the price you face at the point of service is far below the cost of the service, so you use more of it. Mark Pauly named this in a 1968 comment on Arrow, and the term is unfortunate because it sounds like an accusation of bad character. It is not. A person with insurance who fills a prescription they would have skipped is responding to a price, exactly as economics predicts they should.

Distinguish two kinds. Ex ante moral hazard is reduced prevention: if insurance pays for the consequences, you take less care to avoid them. Evidence for this in health is weak, because the consequences of illness are painful regardless of who pays. Ex post moral hazard is increased consumption once ill, and this one is real and large.

The RAND Health Insurance Experiment, running from 1971 to 1982, is still the best measurement anyone has. About 2,750 families were randomly assigned to insurance plans ranging from free care to 95 percent coinsurance, with a cap on annual out-of-pocket spending. Random assignment is what makes it decisive: the groups differed only in the price they faced.

Three findings, all of which matter.

• Price mattered a great deal. Families with free care used roughly a third more care, and spent about 45 percent more, than those facing the highest coinsurance.
• For the average participant, that extra care produced no measurable improvement in health outcomes over the study period.
• For participants who were both poor and sick, free care did help, producing better blood pressure control and corrected vision, with an estimated reduction in mortality risk for that group.

The finding that gets left out of both sides' summaries is the fourth one. Cost sharing reduced clinically appropriate and clinically inappropriate care in roughly equal proportion. Patients facing a price did not skip the useless visit and keep the necessary one; they skipped both, because they cannot reliably tell which is which. That is Arrow's information asymmetry showing up as a measured behavior, and it is the strongest available argument against blunt cost sharing as a cost control tool.

Bottom line: RAND established that patients respond sharply to price, that the average person lost nothing by using less, and that poor and sick people lost real health, because cost sharing suppresses necessary and unnecessary care about equally.

Oregon: what coverage itself does

RAND compared insurance designs among the insured. In 2008 Oregon accidentally ran the other experiment. The state had money to expand Medicaid to a limited number of low-income adults, far fewer than wanted it, so it allocated the slots by lottery. Roughly 90,000 people put their names in and about 30,000 were drawn. A lottery is random assignment, so the winners and losers were comparable in everything except coverage.

The results, published across several papers from 2012 onward, refuse to fit any political summary.

• Coverage increased the use of essentially everything: primary care visits, prescription drugs, and preventive screening.
• It sharply reduced financial strain. Catastrophic out-of-pocket spending nearly disappeared, and medical debt sent to collections fell.
• It produced a large reduction in the rate of depression, on the order of a third in relative terms.
• Self-reported health improved.
• At two years it showed no statistically significant improvement in measured blood pressure, cholesterol, or glycated hemoglobin.
• Emergency department use went up, not down, contradicting the widespread claim that giving people coverage moves them out of the emergency room.

Read the null result carefully rather than triumphantly. Two years is a short window for blood pressure to translate into events, the study was powered to detect fairly large effects, and a null is not proof of no effect. Read the positive results just as carefully: protection from financial ruin is a real benefit of insurance, arguably its primary one, and it does not require any improvement in a biomarker to count.

Notice also what Oregon does to the moral hazard story. The economist John Nyman argues that much of what gets called moral hazard is better described as an income transfer: a sick person who could not previously afford treatment now can, and consumes it. That is not waste. It is the insurance working. Distinguishing the transfer from genuine overuse is hard, and most policy arguments simply assume the answer.

Why this matters: Oregon showed that coverage reliably buys financial protection, mental health, and access, and that its effects on measured physical biomarkers were not detectable in two years, a combination that neither side of the American debate quotes in full.

The cost sharing toolkit

The instruments themselves are simple and worth naming precisely, because plan documents assume you know them.

Instrument	How it works	What it is for
Deductible	You pay the first fixed amount each year before the plan pays anything	Removes small claims from the insurer's administration entirely
Coinsurance	You pay a percentage of the cost after the deductible	Keeps the patient exposed to price even on large claims
Copayment	A flat charge per visit or prescription	Simple and predictable, but unrelated to the actual cost
Out-of-pocket maximum	An annual ceiling after which the plan pays everything in network	Restores the core insurance function against catastrophic loss
Actuarial value	The share of covered costs the plan pays for a standard population	The basis of the marketplace metal tiers you will meet in Module 2

Two practical notes. First, the out-of-pocket maximum is the most important number in any plan document and the one people ignore, because it is the number that decides whether a serious illness is survivable financially. Federal law caps it for most plans, with the ceiling reset annually by regulation. Second, the deductible has grown faster than any other feature of American coverage; the average deductible faced by a covered worker with single employer coverage runs in the region of 1,800 dollars in the mid-2020s, against averages of a few hundred dollars twenty years earlier. That is a large, quiet reduction in the value of coverage that never required anyone to vote on it.

Common misconceptions

• Moral hazard means people are cheating. It names a price response, not dishonesty. A person filling a prescription because it now costs ten dollars is behaving exactly as the model predicts.
• Adverse selection is caused by people lying about their health. It occurs even when everyone is scrupulously honest, because the private information is about likely future use, not about anything one is asked to disclose.
• The RAND experiment proved cost sharing is harmless. It proved it is harmless on average and harmful to poor and sick participants, and that it suppresses necessary and unnecessary care in roughly equal measure.
• Giving people insurance reduces emergency department use. The Oregon lottery found the opposite: emergency use rose. Coverage lowers the price of every setting, including that one.
• You can require insurers to cover everyone at the same price and leave the rest to the market. Guaranteed issue plus community rating without a mandate or subsidy is the exact configuration that produces a death spiral.
• Insurance is a way of paying for routine care. Its economic function is protection against low-probability catastrophic loss; the routine-care role is a policy choice layered on top and financed by everyone's premiums.

Putting it together

• Insurance converts an uncertain large loss into a certain smaller one, and its value rests on the extreme skew of medical spending.
• The actuarially fair premium equals expected claims; real premiums add loading for administration, reserves, and profit.
• Adverse selection arises from the buyer's private knowledge of their own risk and can spiral until only the sickest remain.
• The spiral is blocked only by mandates, subsidies, restricted enrollment, pooling on an unrelated trait, or risk adjustment.
• Harvard's 1995 change to level-dollar subsidies produced a documented spiral that killed its own generous plan within three years.
• RAND showed a large price response, no average health effect, real harm to the poor and sick, and equal suppression of appropriate and inappropriate care.
• Oregon showed that coverage buys utilization, financial protection, and mental health improvement, with no detectable two-year change in three physical biomarkers and higher emergency use.

Sources

1. Cutler, D. M., and Reber, S. J. (1998). Paying for health insurance: The trade-off between competition and adverse selection. Quarterly Journal of Economics, 113(2), 433-466. doi.org
2. Baicker, K., Taubman, S. L., Allen, H. L., Bernstein, M., Gruber, J. H., Newhouse, J. P., Schneider, E. C., Wright, B. J., Zaslavsky, A. M., and Finkelstein, A. N. (2013). The Oregon experiment: Effects of Medicaid on clinical outcomes. New England Journal of Medicine, 368(18), 1713-1722. doi.org
3. Akerlof, G. A. (1970). The market for lemons: Quality uncertainty and the market mechanism. Quarterly Journal of Economics, 84(3), 488-500. doi.org
4. RAND Corporation. (n.d.). The Health Insurance Experiment. rand.org
5. Pauly, M. V. (1968). The economics of moral hazard: Comment. American Economic Review, 58(3), 531-537.

A wartime ruling that nobody meant as health policy

In October 1942 Congress passed the Stabilization Act, which authorized the president to freeze wages in order to hold down inflation while the country fought a war with a labor shortage. Employers who could not compete on salary went looking for something else to offer. In 1943 the National War Labor Board ruled that employer contributions to insurance and pension funds did not count as wages for the purposes of the freeze. The same year, the Internal Revenue Service ruled that those contributions were not taxable income to the employee either. In 1954 Congress wrote the second ruling permanently into the tax code.

Nobody in that sequence was designing a health system. They were managing wartime inflation and then tidying up the tax code. Yet those decisions determine, eighty years later, where roughly 160 million Americans under 65 get their health coverage, why losing a job means losing insurance, why a state legislature cannot fix its own residents' benefits, and why the single largest subsidy in American health care goes disproportionately to people with the highest incomes and never appears as a line in any appropriation.

This lesson follows that accident forward. Treat it as a case study in a general truth about policy: arrangements created for one reason acquire constituencies, and the constituencies outlive the reason.

Key idea: Employer-sponsored insurance is the largest coverage source in the United States and was never chosen; it is the residue of 1940s wage controls and tax rulings, which is why its features look arbitrary when judged as design.

How the exclusion works, in dollars

The mechanism is simple and its consequences are not. If your employer pays you 70,000 dollars in salary and you buy a 20,000 dollar family policy yourself, you pay federal income tax and payroll tax on the full 70,000 and then buy the policy with what is left. If instead your employer pays you 50,000 in salary and 20,000 in premiums, you are taxed only on the 50,000. The 20,000 passes to you untaxed.

Work the size of that gift for two workers.

	Worker in the 12 percent bracket	Worker in the 32 percent bracket
Employer premium contribution	20,000 dollars	20,000 dollars
Federal income tax avoided	2,400 dollars	6,400 dollars
Payroll tax avoided, employee share at 7.65 percent	1,530 dollars	1,530 dollars
Approximate value of the subsidy	3,930 dollars	7,930 dollars

The same policy, the same premium, and roughly double the subsidy for the higher earner. That is not a loophole; it is the arithmetic of excluding income from a progressive tax. Add that lower-income workers are likelier to work for firms that offer no coverage at all, and the exclusion turns out to deliver the most help to the people who need it least. Recent federal estimates put the revenue forgone in the region of 300 billion dollars a year across income and payroll taxes, which makes it the largest single tax expenditure in the budget, larger than the mortgage interest deduction by a wide margin.

Two further consequences follow. Because the subsidy applies to the premium rather than to a fixed amount, it subsidizes generosity: a more expensive plan gets a larger subsidy, which pushes compensation toward richer benefits and away from wages. And because it never appears as spending, it is invisible in budget debates in a way that an equivalent 300 billion dollar appropriation would not be.

So what?: The tax exclusion is an upside-down subsidy, worth roughly twice as much to a high-bracket worker as to a low-bracket one for the identical policy, and it is invisible because it is forgone revenue rather than spending.

Who really pays the premium

Ask an employee who pays for their health insurance and most will say the employer pays most of it. In 2024 the average annual premium for family coverage through an employer was in the region of 25,600 dollars, of which the worker contributed roughly 6,300 through payroll deduction. So the employer pays about three quarters. That is the accounting answer, and economists think it is the wrong answer.

The argument is about incidence, meaning who actually bears a cost as opposed to who writes the check. An employer decides what a worker is worth in total compensation. If health premiums rise by 2,000 dollars and the worker's productivity has not changed, the employer has three options: absorb it as lower profit, pass it to the customer as a higher price, or take it out of what the worker would otherwise have received in wages. In competitive labor markets the evidence points strongly to the third. Jonathan Gruber's study of state laws mandating maternity coverage found that the cost of the mandate showed up almost entirely as lower wages for the affected groups, which is about as clean a test as the question permits.

This matters for how you read the last twenty-five years. Workers frequently report that their pay has stagnated while their employer boasts of generous benefits. Both statements can be true simultaneously, and the relationship between them is causal. The premium growth is coming out of the wage that would otherwise have been paid.

Remember: Employer premium contributions are compensation, not a gift. The evidence on incidence says the worker pays for coverage in the form of wages they never see.

Self-funding and the ERISA wall

Here is the piece that surprises people who have studied insurance regulation and assume states regulate insurance. Most large employers do not buy insurance at all. They self-fund: the employer pays claims out of its own money, hires an insurance company as an administrator to process them, and buys stop-loss coverage against a catastrophic year. Roughly six in ten covered workers are in such an arrangement, and at large firms the great majority are.

The reason is the Employee Retirement Income Security Act of 1974. ERISA was passed after a decade of pension failures and was mostly about retirement security, but it contains a preemption clause that supersedes state laws relating to employee benefit plans. A savings clause preserves state authority to regulate insurance. And then a deemer clause forbids states from deeming a self-funded plan to be an insurer. Put the three together and you get the structure of American benefits law: a state may regulate an insurance policy sold to an employer, and may not regulate the self-funded plan next door that provides identical benefits.

	Fully insured plan	Self-funded plan
Who bears the claims risk	The insurer	The employer, above a stop-loss threshold
State benefit mandates apply	Yes	No
State premium taxes apply	Yes	No
State external review and appeal rules apply	Yes	Generally no; federal rules apply instead
Typical adopter	Small and mid-sized firms	Large employers and, increasingly, mid-sized ones

The practical effect is that a state that wants to require, say, coverage of a particular therapy can reach only part of its own workforce. It is also why proposals for state-level universal coverage keep running into a legal wall, and why the serious versions require a federal waiver of ERISA that Congress has never granted. The Supreme Court has policed the boundary repeatedly, allowing states to tax hospital rates and to regulate the reimbursement practices of pharmacy benefit managers, while striking down state attempts to reach into plan design.

What matters here: ERISA preemption plus self-funding removes most large-employer coverage from state jurisdiction, which is the single largest legal obstacle to state-level health reform.

Job lock, and how big it actually is

If coverage comes with the job, then leaving the job costs you coverage, and people with a sick family member may stay in work they would otherwise leave. That is job lock, and it is the most cited social cost of the employer system.

The evidence is more equivocal than the rhetoric. Brigitte Madrian's 1994 study compared job mobility for workers with and without alternative coverage sources and estimated that employer coverage reduced voluntary job turnover substantially, on the order of a quarter. Later work using different designs found smaller effects, and the estimates remain a live disagreement in labor economics rather than a settled number. What is not in dispute is the direction and the mechanism, and researchers have since found related effects on entrepreneurship: people are more likely to start a business when they have coverage from another source, such as a spouse or, after 65, Medicare.

Two federal statutes chip at the problem without solving it. COBRA, enacted in 1985, lets a departing employee continue the group plan for a period, but at the full premium plus an administrative charge, which is exactly when the person has no paycheck. HIPAA in 1996 limited preexisting condition exclusions when moving between group plans. The ACA's marketplaces did more than either, by creating a subsidized place to buy coverage that is not attached to a job, and studies of self-employment and retirement timing after 2014 found measurable effects.

What is actually wrong with it, and what is not

It is easy to write a list of complaints about employer coverage and harder to be fair about it, so hold both columns.

In its favor: it pools risk on a trait, employment, that is only weakly related to health, which is a genuinely elegant solution to adverse selection and one that no one designed. Large employers are informed, motivated purchasers with real bargaining leverage. Enrollment is automatic and take-up is high. And the coverage is generally more generous than what individuals buy for themselves.

Against it: the tax subsidy is regressive and hidden. Coverage is tied to a job, so it disappears exactly when income does, in a recession or an illness that stops you working. Small employers face far higher administrative costs per worker and much lower offer rates, so the system's benefits are concentrated among people who already have stable, well paid work. Job lock is real even if its magnitude is argued. And the arrangement burdens firms with a function, purchasing medical care, that has nothing to do with their business, which is why American firms competing internationally periodically complain about it.

One more, easy to miss: because employers buy on behalf of a captive group, the price signal that is supposed to discipline health care spending reaches almost nobody. The employee sees a payroll deduction, not a premium. The employer sees a premium but not the prices inside it. That opacity is a precondition for the price variation you will study in Module 4.

The upshot: Employer coverage solves the selection problem elegantly and distributes its subsidy backwards, ties coverage to the thing most likely to disappear during illness, and hides prices from everyone who might respond to them.

Common misconceptions

• Employers provide health insurance out of generosity or because the law requires it. Most do it because it is tax-favored compensation, and the ACA employer requirement applies only to larger firms and is modest in force.
• The employer pays most of the premium. Accounting says yes; incidence evidence says the cost comes out of wages, so the worker pays for nearly all of it.
• States regulate the health insurance their residents have. States regulate fully insured policies. ERISA and the deemer clause put self-funded plans, covering most large-firm workers, beyond their reach.
• Self-funding means the employer bought a cheaper insurance policy. It means the employer is the insurer, buying only administration and catastrophic stop-loss protection.
• The tax exclusion helps ordinary workers most. Its value rises with the marginal tax rate, so it is worth roughly twice as much to a high earner as to a low earner for the identical policy.
• COBRA protects people who lose their jobs. It preserves the right to buy the same plan at the full unsubsidized premium, which is why take-up among the newly unemployed is low.

Looking back

• Employer coverage originated in the 1942 wage freeze, the 1943 War Labor Board and IRS rulings, and the 1954 codification of the tax exclusion.
• It now covers roughly 160 million Americans under 65, more than any other source.
• The exclusion is the largest federal tax expenditure, in the region of 300 billion dollars a year, and is worth more to higher-bracket taxpayers.
• Average family premiums ran around 25,600 dollars in 2024, with the worker contributing roughly 6,300 directly and, on the incidence evidence, the rest through foregone wages.
• ERISA preemption plus the deemer clause exempts self-funded plans from state insurance regulation, which is the chief legal barrier to state-level reform.
• Job lock is real and contested in magnitude; COBRA, HIPAA, and the ACA marketplaces each reduce it partially.

Sources

1. KFF. (n.d.). Employer Health Benefits Survey. kff.org
2. Congressional Budget Office. (n.d.). Taxes. cbo.gov
3. United States Department of Labor, Employee Benefits Security Administration. (n.d.). Employee Retirement Income Security Act (ERISA). dol.gov
4. Madrian, B. C. (1994). Employment-based health insurance and job mobility: Is there evidence of job-lock? Quarterly Journal of Economics, 109(1), 27-54. doi.org
5. Gruber, J. (1994). The incidence of mandated maternity benefits. American Economic Review, 84(3), 622-641.

Three bills stapled together

On 30 July 1965 Lyndon Johnson signed the Social Security Amendments in the Harry S. Truman Library in Independence, Missouri, and handed the first Medicare card to the 81-year-old Truman, who had asked Congress for national health insurance in 1945 and been refused. The scene was arranged for the symbolism. What actually passed that day was stranger than the symbolism suggests.

Three rival proposals had been in play. The administration wanted compulsory hospital insurance for the elderly, financed by payroll tax. Republicans, led by John Byrnes, countered with a voluntary program covering physician services, financed by premiums and general revenue. The American Medical Association, opposed to both, proposed instead to expand an existing welfare program that paid medical bills for the indigent elderly. Wilbur Mills, chairman of the House Ways and Means Committee and until then an obstacle to any of it, did something nobody expected: he took all three and enacted them at once.

The hospital plan became Medicare Part A. The Republican physician plan became Medicare Part B. The welfare expansion became Medicaid, a separate title of the Social Security Act with a separate philosophy. That is why the two programs, born the same afternoon, share almost nothing structurally. Medicare is federal, uniform across states, financed by payroll tax and general revenue, and available to anyone who reaches 65. Medicaid is jointly financed and state-administered, varies enormously across state lines, and is means-tested. Learning them as a pair, and keeping their differences straight, is the single most useful piece of American health policy knowledge you can acquire.

Key idea: Medicare and Medicaid differ so profoundly because they were three separate bills combined for legislative convenience in 1965, not two halves of one design.

Medicare, part by part

Medicare covers people aged 65 and over, people under 65 who have received Social Security disability benefits for 24 months, and people with end-stage renal disease or amyotrophic lateral sclerosis. Enrollment is now in the region of 68 million people.

Part	What it covers	How it is paid for	Created
A, Hospital Insurance	Inpatient hospital, skilled nursing after a qualifying stay, hospice, some home health	A dedicated 2.9 percent payroll tax split between worker and employer, held in a trust fund	1965
B, Supplementary Medical Insurance	Physician services, outpatient care, durable equipment, most preventive services	About three quarters general federal revenue, about one quarter beneficiary premiums	1965
C, Medicare Advantage	Parts A and B, usually D, delivered through a private plan under contract	Capitated payments from Medicare, risk-adjusted per enrollee	1997, renamed 2003
D, Prescription Drugs	Outpatient prescription drugs through private plans	General revenue, beneficiary premiums, state contributions for dual eligibles	2003, effective 2006

Two features of that table deserve more attention than they usually get. First, Part A is the only part financed by a dedicated tax with its own trust fund, which is why the annual Trustees report generates headlines about Medicare running out of money. The projected depletion date has moved repeatedly, sitting in the 2030s in recent reports, and depletion would mean incoming payroll tax revenue covering most but not all of Part A benefits, not the program stopping. Parts B and D draw on general revenue and cannot be depleted in that sense; they simply grow as a claim on the federal budget.

Second, and more important for anyone actually enrolled: traditional Medicare has no annual cap on out-of-pocket spending. A beneficiary facing a catastrophic year has unlimited exposure to coinsurance. That single design gap, an insurance program without the feature that makes insurance worth buying, is why a whole secondary market exists. Roughly speaking, a person in traditional Medicare needs either a Medigap policy, retiree coverage from a former employer, Medicaid, or Medicare Advantage in order to be genuinely protected.

Why this matters: Traditional Medicare has no out-of-pocket maximum, so the supplemental market is not a luxury add-on but a repair to the program's central gap.

What Medicare does not cover

Ask people what Medicare covers and most will say everything. The omissions are large, consequential, and follow from 1965 assumptions that nobody has revisited wholesale.

• Long-term custodial care, meaning help with bathing, dressing, and eating over months or years, is not covered. Medicare pays for skilled nursing only after a qualifying hospital stay and only for a limited period. This is the single largest misunderstanding in American health policy, and families discover it at the worst possible moment.
• Routine dental, vision, and hearing services are excluded from traditional Medicare. Many Medicare Advantage plans offer limited versions of these as supplemental benefits, financed out of rebate dollars, which is a large part of why enrollment has moved.
• Care outside the United States is generally not covered.

The long-term care gap is the one to remember, because it is what pushes middle-class families into Medicaid. A person who needs nursing home care spends down their assets until they qualify for Medicaid, which does cover it. That is why Medicaid, a program conceived as welfare medicine for the poor, is the largest payer for nursing home care in the country.

Medicare Advantage, and the argument about it

In 2023 the share of Medicare beneficiaries enrolled in private Medicare Advantage plans passed half for the first time, and has continued rising. That is a structural change in the largest health program in the country, and it happened without any single piece of legislation announcing it.

The mechanics: a plan bids against a county benchmark. If it bids below the benchmark it keeps a share of the difference as a rebate, which must be spent on extra benefits, reduced cost sharing, or lower premiums. Payments are risk-adjusted using diagnosis codes, so a plan enrolling sicker members receives more. Plans are also rated on a five-star quality scale, and high ratings bring bonus payments.

The case for it is real. Beneficiaries get an out-of-pocket maximum, which traditional Medicare lacks, plus dental, vision, hearing, and sometimes fitness benefits, often at zero additional premium. Plans manage care, and their utilization of some services is lower than in traditional Medicare.

The case against is also real, and it is mainly about the payment formula. Because payment rises with documented diagnoses, plans have a direct financial incentive to document more diagnoses, through chart reviews and in-home assessments. Whether that reflects better detection or coding intensity is the central empirical dispute. The Medicare Payment Advisory Commission, which advises Congress, has concluded repeatedly that Medicare pays substantially more for Medicare Advantage enrollees, on the order of twenty percent more, than traditional Medicare would have spent on comparable people, amounting to tens of billions of dollars a year. Set against that are documented complaints about prior authorization denials and narrow networks.

The point: Medicare Advantage now covers most beneficiaries and gives them protections traditional Medicare lacks, while a payment formula tied to diagnosis coding has made it cost Medicare more per person rather than less.

Medicaid: a different animal entirely

Medicaid is not a federal program with state offices. It is fifty-six programs, one per state and territory plus the District of Columbia, operating within federal rules, each with its own eligibility levels, benefit choices, and payment rates. Understanding it means holding two facts at once: it is an individual entitlement, so a person who qualifies must be enrolled and served, and it is an open-ended federal match, so a state that spends more receives more.

The match is called the Federal Medical Assistance Percentage. It is calculated from state per capita income, ranging from a floor of 50 percent in the wealthiest states to over 70 percent in the poorest. For the adults covered under the ACA expansion, the federal share is 90 percent, a deliberately generous rate designed to make expansion attractive.

Federal law then divides both eligibility and benefits into mandatory and optional. States must cover certain groups, including low-income children, pregnant women, and very poor parents, and must provide services including inpatient and outpatient hospital care, physician services, laboratory and imaging, and, for children, the comprehensive Early and Periodic Screening, Diagnostic and Treatment benefit. Beyond that, states choose: prescription drugs are technically optional and every state covers them; dental for adults, physical therapy, and home and community-based services vary enormously.

Here is the scale of the thing, and it surprises people who think of Medicaid as a small welfare program. Enrollment in Medicaid and CHIP peaked at roughly 94 million people in spring 2023, inflated by a pandemic-era rule that barred states from disenrolling anyone, and then fell substantially through the process known as the unwinding when that rule expired. Medicaid finances roughly four in ten births in the United States. It covers roughly six in ten nursing facility residents. It is the largest single payer for behavioral health services. And through the Children's Health Insurance Program, created in 1997, it extends to children in families earning too much for Medicaid but too little to buy coverage.

Bottom line: Medicaid is a state-administered entitlement with an open-ended federal match, covering a far broader population and a far wider set of services, especially long-term care, than its welfare-program reputation suggests.

Where the two programs meet

About twelve million people qualify for both. Dual eligibles are, as a group, older, poorer, sicker, and far more likely to need long-term services than either program's typical enrollee, and they account for a disproportionate share of both programs' spending. Medicare pays first for acute care; Medicaid picks up premiums, cost sharing, and the long-term care Medicare does not cover.

The result is a coordination failure with a name: cost shifting between the two payers. If a nursing home resident's condition deteriorates, hospitalizing them moves the cost from Medicaid, the state's budget, to Medicare, the federal one. The incentive is obvious, the effect is documented, and successive integration programs have tried to fix it by giving one entity responsibility for the whole person. That is worth knowing as a general lesson: whenever two payers cover different pieces of one patient, expect the boundary to be exploited, not out of malice but because budgets are separate.

Common misconceptions

• Medicare covers nursing homes. It covers limited skilled nursing after a qualifying hospital stay. Custodial long-term care is the province of Medicaid, reached by spending down assets.
• Medicare is free because you paid in. The payroll tax funds Part A only. Parts B and D carry monthly premiums, deductibles, and coinsurance, and higher earners pay income-related surcharges.
• Medicare will go bankrupt. Depletion of the Part A trust fund would mean incoming payroll revenue covering most but not all Part A benefits. Parts B and D are financed from general revenue and cannot be depleted in that sense.
• Medicaid is a program for people who do not work. Most adult Medicaid enrollees who can work do, typically in jobs that offer no coverage. A large share of enrollees are children, older adults, and people with disabilities.
• Medicaid is the same everywhere. Eligibility levels, optional benefits, and provider payment rates vary so much between states that the program a person receives depends heavily on where they live.
• Medicare Advantage saves Medicare money. It was designed to, and on the Medicare Payment Advisory Commission's analysis it now costs the program more per beneficiary than traditional Medicare would have.

The short version

• The 1965 Social Security Amendments combined three rival proposals: hospital insurance, voluntary physician coverage, and an expanded welfare medical program, producing Parts A and B of Medicare and Medicaid.
• Medicare is federal and uniform, covering people 65 and over, long-term disability recipients, and those with end-stage renal disease or ALS, with enrollment around 68 million.
• Part A is payroll-tax financed through a trust fund; Parts B and D draw mainly on general revenue and premiums; Part C delivers all of it through private plans.
• Traditional Medicare has no out-of-pocket maximum and excludes long-term custodial care, routine dental, vision, and hearing.
• More than half of beneficiaries are now in Medicare Advantage, which supplies the missing out-of-pocket cap while costing Medicare more per person on MedPAC's analysis.
• Medicaid is a state-administered entitlement with a federal match from 50 percent to over 70 percent, and 90 percent for the ACA expansion population.
• Medicaid and CHIP enrollment peaked near 94 million in 2023 and then fell during the unwinding; the program finances about four in ten births and covers about six in ten nursing facility residents.

Sources

1. Social Security Administration. (n.d.). History: Social Security Amendments of 1965. ssa.gov
2. Centers for Medicare and Medicaid Services. (n.d.). Centers for Medicare and Medicaid Services. cms.gov
3. Medicaid.gov. (n.d.). Medicaid. medicaid.gov
4. Medicare Payment Advisory Commission. (n.d.). Reports to the Congress. medpac.gov
5. KFF. (n.d.). Medicare. kff.org

The number the law was written to move

In 2010 about 48 million people in the United States, roughly one in six, had no health insurance. By 2023 the figure was around 25 million and the uninsured rate had fallen to about 7.6 percent, the lowest ever recorded in the National Health Interview Survey. Marketplace enrollment, which was zero before 2014, reached about 21 million people for 2024 and about 24 million for 2025.

Those are the outcome numbers, and they are the fairest place to start, because arguments about the Patient Protection and Affordable Care Act tend to be conducted entirely in terms of intentions. The law was signed on 23 March 2010, amended a week later by a reconciliation bill, and has since survived two Supreme Court challenges on its merits, one on standing, a repeal vote that failed by one, and the removal of one of its central provisions. It is worth understanding what it actually did, which is both more and less than either its supporters or its opponents usually say.

What matters here: The ACA cut the uninsured rate by more than half without changing the fundamental structure of American health care, which is simultaneously its achievement and its limit.

Why the three pieces had to travel together

Start from the problem you learned in Lesson 2. Before 2014, an insurer selling in the individual market could refuse to sell to a person with a preexisting condition, could charge more based on health status, and could exclude a condition from coverage. Those practices are indefensible from the point of view of the person turned away and entirely rational from the insurer's point of view: an insurer that accepted everyone at a common price, while its competitors did not, would attract the sick and go under.

So the law banned them. Guaranteed issue requires insurers to sell to anyone. Modified community rating allows premiums to vary only by age, within a three-to-one band, by tobacco use, by geography, and by family size, and forbids variation by sex or health status.

You already know what happens next. Those two rules on their own are the death spiral recipe. The law's answer was a three-legged stool.

Leg	Provision	Job it does
Insurance rules	Guaranteed issue, modified community rating, no preexisting condition exclusions, essential health benefits	Makes coverage available and comparable
Individual mandate	A tax penalty for going without qualifying coverage	Keeps healthy people in the pool
Subsidies	Premium tax credits on a sliding scale, plus cost-sharing reductions for lower-income silver enrollees	Makes buying affordable, and therefore rational, at low incomes

Then one leg was removed. The 2017 tax law set the mandate penalty to zero from 2019, and the predicted collapse did not happen. Enrollment dipped and then grew. The lesson analysts drew is that the subsidy leg was doing far more work than the mandate leg: for a person receiving a large premium tax credit, buying is worth doing regardless of any penalty. The mandate was politically the most visible piece and economically among the least important, which is a useful warning about reading a statute by its headlines.

The subsidy leg was then strengthened. Legislation in 2021 and 2022 increased the credits and removed the hard cutoff at four times the federal poverty level, capping premium contributions as a share of income instead. Marketplace enrollment rose sharply afterward, which is about as direct a demonstration of subsidy elasticity as policy provides. Those enhancements were enacted with expiration dates, so the enrollment figures you look up will depend on what Congress did about them, and you should check the current status rather than trusting any textbook number.

In short: The stool's three legs are not equally load-bearing. Removing the mandate barely moved enrollment; changing the subsidies moved it a great deal.

How a marketplace plan is actually priced

Work one household through it, because the mechanics are where the arguments live.

Plans are sorted into metal tiers by actuarial value, the share of covered costs the plan pays for a standard population: bronze about 60 percent, silver about 70, gold about 80, platinum about 90. In each rating area, the second-lowest-cost silver plan is designated the benchmark. The premium tax credit is calculated so that a household's contribution toward that benchmark plan does not exceed a set percentage of its income, and the credit equals the difference. Crucially, the household may apply that credit to any metal tier, so a family that takes a bronze plan often pays little or nothing in premium while accepting a much larger deductible.

Cost-sharing reductions work differently and are frequently misunderstood. For households below 250 percent of the federal poverty level, a silver plan's deductibles and coinsurance are reduced, sometimes dramatically, raising its effective actuarial value well above 70 percent. Those reductions are available only on silver, which is why the standard advice for a lower-income shopper is to look at silver first even when a bronze premium looks cheaper.

The law also imposed a medical loss ratio requirement: insurers must spend at least 80 percent of premium revenue in the individual and small group markets, and 85 percent in the large group market, on claims and quality improvement, and rebate the difference. That is a direct cap on the loading you met in Lesson 2, and it has returned billions of dollars in rebates.

The gap the Supreme Court created

As drafted, the law extended Medicaid to everyone under 138 percent of the federal poverty level, with the federal government paying 90 percent of the cost, and it enforced this by threatening a state's entire Medicaid funding if it refused. In National Federation of Independent Business v. Sebelius, decided in 2012, the Supreme Court upheld the individual mandate as a tax but held that the Medicaid enforcement mechanism was unconstitutionally coercive. Expansion became optional.

The consequence is a hole that no one designed and that is easy to state precisely. The drafters assumed everyone below 100 percent of poverty would be on Medicaid, so they wrote marketplace subsidies to begin at 100 percent. In a state that did not expand, an adult below 100 percent of poverty is therefore often too poor for a subsidy and too well off for that state's Medicaid, which in some states covers non-disabled childless adults not at all.

Picture the person. A 40-year-old with no children earning 13,000 dollars a year in a non-expansion state. In an expansion state she is on Medicaid with no premium. In a non-expansion state she is in the coverage gap: ineligible for Medicaid because she has no qualifying category, ineligible for a premium tax credit because she earns too little. Someone earning twice as much qualifies for substantial help; she qualifies for none. As of the mid-2020s, 40 states and the District of Columbia had adopted expansion, and ten had not, so the gap is now a regional phenomenon concentrated in the South.

So what?: The coverage gap is a legal artifact, not a policy choice: it exists because a court made optional a provision the subsidy formula assumed would be universal.

What the law did not change

This is the section most summaries skip, and it explains why American health care in 2026 feels much as it did in 2009 to anyone with employer coverage.

• It left the employer system intact. The largest single source of coverage was barely touched, apart from a requirement that larger employers offer coverage or pay a penalty, and the excise tax on high-cost plans, which was repealed in 2019 before ever taking effect.
• It did almost nothing about prices. The law regulated insurance and expanded coverage. It did not set or limit what hospitals, physicians, or drug manufacturers may charge commercial payers, which is where American spending diverges from everyone else's.
• It did not simplify anything. A person's coverage still depends on age, income, employer, state, and immigration status, and the law added a new category with its own eligibility rules and its own annual enrollment ritual.
• It did not achieve universal coverage and did not claim to. The Congressional Budget Office always projected millions remaining uninsured, and undocumented immigrants were excluded from the marketplaces by statute.
• It changed provider payment only experimentally. Accountable care organizations, readmission penalties, and the Innovation Center were created by the law, and Module 3 assesses how much they actually moved.

Who is still uninsured, and why

The composition of the remaining uninsured is more surprising than the number. The great majority live in families with at least one worker. A large share, roughly six in ten of uninsured nonelderly people by KFF's analysis, are eligible for Medicaid or for subsidized marketplace coverage and are not enrolled, for reasons that include not knowing they qualify, finding the process difficult, or having tried and lost coverage during a redetermination.

The rest divide into groups the law cannot reach as written. People in the coverage gap in non-expansion states. Noncitizens, including lawfully present immigrants in a five-year waiting period for Medicaid and undocumented residents excluded entirely, although several states now cover some of them with state funds. And people who are eligible for a subsidy but still find the premium or the deductible unaffordable given everything else they owe.

The policy implication is worth stating because it is counterintuitive. Enrollment failure, not eligibility failure, is now the largest single component of American uninsurance. That makes outreach, automatic enrollment, and simpler renewal processes more powerful levers than they sound, and it is why the unwinding of pandemic-era Medicaid rules mattered so much: it was an enrollment event, not an eligibility change, and it moved millions of people.

Common misconceptions

• The ACA created government-run health care. It expanded a means-tested public program and subsidized the purchase of private insurance. No provider was nationalized and no new public insurer was created.
• Removing the individual mandate destroyed the marketplaces. Enrollment dipped and then grew. The subsidy leg turned out to be doing most of the work.
• Everyone below the poverty line is covered by Medicaid. In the ten states that have not expanded, many poor adults qualify for neither Medicaid nor a subsidy, which is the coverage gap.
• Bronze plans are the best deal for low-income shoppers. Cost-sharing reductions attach only to silver plans, so silver is frequently the better value below 250 percent of poverty despite the higher sticker premium.
• The uninsured are people who do not work. Most live in working families, and most are eligible for help they are not receiving.
• The law addressed health care costs. It addressed coverage and insurance market conduct. Prices, the main driver of American spending, were left essentially untouched.

What to carry forward

• The uninsured rate fell from roughly one in six in 2010 to about 7.6 percent in 2023, with marketplace enrollment reaching around 24 million for 2025.
• Guaranteed issue and community rating require a mandate or subsidies to avoid a death spiral; the ACA supplied both, and only the subsidies proved essential.
• Premium tax credits are pegged to the second-lowest-cost silver plan and may be applied to any tier; cost-sharing reductions attach only to silver.
• Medical loss ratio rules cap insurer loading at 20 percent in the individual and small group markets and 15 percent in the large group market.
• NFIB v. Sebelius made Medicaid expansion optional, producing the coverage gap for poor adults in the ten states that have not adopted it.
• The law left employer coverage, provider prices, and the overall structure of the system substantially unchanged.
• Most remaining uninsured people are in working families, and most are eligible for coverage they are not enrolled in.

Sources

1. KFF. (n.d.). Affordable Care Act. kff.org
2. HealthCare.gov. (n.d.). Health insurance marketplace. healthcare.gov
3. Congressional Budget Office. (n.d.). Health Care. cbo.gov
4. National Center for Health Statistics. (n.d.). National Health Interview Survey. cdc.gov
5. National Federation of Independent Business v. Sebelius, 567 U.S. 519 (2012).

The law that built the hospitals

The Hospital Survey and Construction Act of 1946, known as Hill-Burton after its Senate sponsors, gave federal money to states to build hospitals in places that had none. Recipients took on obligations in return: to provide a reasonable volume of free care to those unable to pay, and to make services available to everyone in the area they served. Over roughly two decades it helped finance construction across thousands of communities and produced much of the physical hospital stock the country still uses.

It also contained a clause permitting separate facilities for different races so long as they were of equal quality, and in practice that meant segregated hospitals built with federal money. A group of Black physicians and dentists in Greensboro, North Carolina sued, and in 1963 the Fourth Circuit held in Simkins v. Moses H. Cone Memorial Hospital that the provision was unconstitutional. Two years later Medicare arrived, and hospitals that wanted its money had to desegregate to be certified. That is one of the clearest cases anywhere of payment policy accomplishing what civil rights litigation alone had not.

Both halves of that story are worth carrying. The American hospital sector is not a market outcome. It was built with public money under public conditions, and it is still shaped by the conditions attached to public payment.

Key idea: The hospital system was constructed with federal subsidy under federal conditions, and Medicare certification, not litigation alone, is what finally desegregated it.

What the hospital sector looks like now

There are roughly 6,100 hospitals in the United States. Of the community hospitals among them, about three in five are private nonprofit, about one in four are for-profit, and the rest are owned by state or local government. That ownership mix is unusual internationally and it matters for how the sector behaves.

A nonprofit hospital pays no federal income tax, no state or local property tax in most jurisdictions, and can issue tax-exempt bonds. In exchange it is expected to provide community benefit. Until 2010 that expectation was vague. The ACA added requirements now found in section 501(r) of the tax code: a nonprofit hospital must conduct a community health needs assessment every three years and adopt a strategy responding to it, must have a written financial assistance policy and publicize it, must limit what it charges patients eligible for assistance to the amounts generally billed to insured patients, and must make reasonable efforts to determine eligibility before pursuing extraordinary collection actions such as lawsuits or wage garnishment.

Whether those rules did much is genuinely disputed. Investigative reporting has repeatedly found nonprofit systems suing patients who would have qualified for free care under the hospital's own policy, and studies comparing the value of the tax exemption with the charity care provided find wide variation, with some systems clearly ahead and others clearly behind. The rules made the policies public; they did not make them generous.

One more statute defines the sector. EMTALA, enacted in 1986, requires any hospital with an emergency department that participates in Medicare, which is nearly all of them, to provide a medical screening examination to anyone who comes in, and to stabilize an emergency condition before transfer or discharge, regardless of ability to pay. This is the closest thing the United States has to a universal right to care, and Congress attached no money to it. That combination, a legal duty to treat with no financing, is why emergency departments function as the safety net of last resort and why uncompensated care is concentrated in hospitals serving the poorest areas.

What matters here: EMTALA created a universal duty to screen and stabilize without creating any way to pay for it, which is why the emergency department is where American coverage gaps become visible.

Consolidation and what the evidence shows

Over the past three decades hospitals have merged at a steady rate, and independent physician practices have been bought by hospitals and, increasingly, by insurers and private equity firms. By the mid-2020s roughly three quarters of American physicians were employed by hospitals or other corporate entities rather than working in practices they owned.

The stated rationale is coordination: an integrated system can move a patient from clinic to hospital to rehabilitation without dropping the chart. The evidence on what actually happens is unusually clear for a health policy question, because prices are observable in claims data.

• Zack Cooper and colleagues, using a large national claims database, found that hospital prices in markets with a single hospital were roughly 12 percent higher than in markets with four or more, after adjusting for the mix of services.
• Studies of completed mergers find price increases concentrated where the merging hospitals were close substitutes for one another, which is exactly what antitrust theory predicts.
• Research on hospitals acquiring physician practices finds prices for those physicians' services rise afterward, partly because a clinic owned by a hospital can bill a facility fee that an independent clinic cannot.
• Quality gains have not shown up. Work comparing acquired hospitals with matched controls found no improvement in mortality or readmissions and modest worsening in patient experience scores.

That combination, higher prices without measurable quality gains, is the strongest empirical case in American health policy for more aggressive antitrust enforcement, and it is worth noting that it does not depend on anyone behaving badly. A larger system negotiating with an insurer that must have it in the network will get a better price. That is what bargaining power is.

The upshot: Hospital and physician consolidation raises prices reliably and improves measured quality rarely, which makes it one of the few health policy questions where the evidence points in a single direction.

Where physicians come from

The supply of physicians is not set by demand. It is set by a pipeline with several deliberate constrictions, and you cannot understand shortage debates without knowing where they are.

Stage	Typical duration	What limits capacity
Undergraduate degree	4 years	Nothing specific to medicine
Medical school, MD or DO	4 years	Accredited class sizes, which have grown substantially since 2000
Residency	3 to 7 years	Funded positions, capped for Medicare purposes by the Balanced Budget Act of 1997
Fellowship, optional	1 to 3 years	Program capacity by subspecialty
Licensure and board certification	Ongoing	State medical boards; specialty boards

The binding constraint is the third row. Medicare is the largest funder of graduate medical education, and in 1997 Congress capped the number of residency positions it would support at each hospital, on the then-prevailing view that the country had too many physicians. Medical schools subsequently expanded enrollment considerably, but a medical school graduate without a residency position cannot practice. Congress has added slots since, in increments of hundreds rather than thousands. The Association of American Medical Colleges has projected a shortfall reaching tens of thousands of physicians within the next decade, with the largest gaps in primary care and in non-metropolitan areas.

The distribution problem is at least as serious as the total. The Health Resources and Services Administration designates health professional shortage areas, and there are thousands of them, rural and urban both. Payment explains much of it: a procedural specialist in a metropolitan area with a favorable payer mix earns multiples of what a primary care physician in a rural county earns, and medical students carrying large debts respond to that. The National Health Service Corps, which repays loans in exchange for service in shortage areas, is the main federal counterweight and is small relative to the gap.

Nurses, and the staffing evidence

Registered nurses are the largest health profession in the country, roughly 3.2 million strong, and nurse staffing is one of the few structural variables with a solid, replicated link to patient outcomes.

The landmark study is Linda Aiken and colleagues in 2002, which linked nurse staffing data from Pennsylvania hospitals to surgical patient outcomes and found that each additional patient per nurse was associated with roughly a 7 percent increase in the odds of dying within 30 days of admission, and a similar increase in the odds of failure to rescue, meaning death following a complication. The mechanism is intuitive: complications are usually survivable when noticed early, and noticing requires someone at the bedside.

California acted on this before the evidence was complete. A 1999 law directed the state to set minimum nurse-to-patient ratios by unit type, and the ratios took effect in 2004. Subsequent comparisons of California with states lacking ratios found lower nurse workloads and better outcomes, though disentangling the ratio law from everything else that differs between states is hard, and hospitals argued the mandate was expensive and inflexible. Several other states have adopted disclosure or committee requirements rather than fixed ratios.

Alongside nurses sit nurse practitioners and physician assistants, whose numbers have grown far faster than the physician workforce. What they may do without physician supervision is set state by state, and roughly half the states now grant nurse practitioners full practice authority. Studies comparing primary care outcomes across these arrangements have generally found comparable quality for the conditions studied, and the dispute is less about the studies than about which conditions and which patients they cover.

Remember: Nurse staffing is one of the few inputs with a replicated association with mortality, which is why it is regulated in some states and fought over in all of them.

Common misconceptions

• Nonprofit hospitals do not seek profit. They seek revenue over expenses, which is called a margin rather than a profit and is required to fund construction and equipment. What distinguishes them is that no one owns the surplus, not that they do not pursue one.
• EMTALA guarantees people health care. It guarantees a screening examination and stabilization of an emergency condition. It does not cover ongoing treatment, and it does not stop the hospital billing afterward.
• Hospital mergers lower costs through efficiency. Measured prices rise, and quality improvements have not appeared in the comparative studies.
• The physician shortage is caused by too few medical school places. Enrollment has grown substantially; the binding constraint is funded residency positions, capped since 1997.
• Nurse staffing ratios are simply a labor demand. They originate in a body of outcomes research linking workload to mortality, which does not settle the policy question but does mean the debate is not only about wages.
• The safety net is a set of public hospitals. It is mostly a legal duty imposed on all hospitals with emergency departments, financed by cross-subsidy from other patients.

Pulling it together

• Hill-Burton built much of the hospital stock from 1946 with conditions attached, including a segregation clause struck down in 1963, and Medicare certification then forced desegregation.
• Roughly 6,100 hospitals operate in the United States, with community hospitals about three fifths nonprofit, a quarter for-profit, and the remainder governmental.
• Section 501(r) requires community health needs assessments, published financial assistance policies, and limits on collection actions, with contested effects.
• EMTALA creates a duty to screen and stabilize with no financing attached.
• Consolidation raises prices, with monopoly-market hospital prices roughly 12 percent above competitive ones, and has not produced measured quality gains.
• Physician supply is limited chiefly by residency positions capped in 1997, and maldistributed by payment differences across specialty and geography.
• Each additional patient per nurse is associated with about a 7 percent increase in 30-day mortality odds, the finding behind California's ratio law.

Sources

1. American Hospital Association. (n.d.). Fast Facts on U.S. Hospitals. aha.org
2. Aiken, L. H., Clarke, S. P., Sloane, D. M., Sochalski, J., and Silber, J. H. (2002). Hospital nurse staffing and patient mortality, nurse burnout, and job dissatisfaction. JAMA, 288(16), 1987-1993. doi.org
3. Cooper, Z., Craig, S. V., Gaynor, M., and Van Reenen, J. (2019). The price ain't right? Hospital prices and health spending on the privately insured. Quarterly Journal of Economics, 134(1), 51-107. doi.org
4. Bureau of Labor Statistics. (n.d.). Occupational Outlook Handbook: Registered Nurses. bls.gov
5. Association of American Medical Colleges. (n.d.). Workforce Data and Reports. aamc.org

The day the incentive flipped

On 1 October 1983 Medicare stopped paying hospitals for what a patient's stay had cost and began paying a fixed amount per admission, set in advance according to the patient's diagnosis. A hospital that sent a pneumonia patient home on day four and a hospital that kept an identical patient until day nine now received the same payment. Overnight, an extra day changed from revenue into cost.

Hospitals responded exactly as anyone would predict. Lengths of stay fell sharply within two years. Researchers immediately asked the obvious follow-up question, which entered the literature under the phrase quicker and sicker: were patients being discharged before they were ready? The careful studies found that patients were indeed leaving in less stable condition, while outcomes overall did not deteriorate as much as critics feared, partly because care moved into skilled nursing facilities and home health rather than disappearing.

Hold that whole episode in mind, because it is the template for this lesson. Change what you pay for and behavior changes fast. Some of the change is what you wanted. Some of it is the system finding the seam in your rule. There is no payment method without a seam, and choosing a payment method is choosing which distortion you are willing to live with.

Key idea: Every payment method produces a predictable distortion. The policy question is never how to eliminate the distortion but which one is least harmful and how to blunt it.

The spectrum, in one table

Order payment methods by the size of the unit being paid for, from a single service up to a whole population for a whole year. As the unit grows, financial risk moves from payer to provider, and the direction of the distortion reverses.

Method	Unit of payment	Who bears risk	Characteristic distortion
Fee-for-service	One service	Payer	Too much volume, especially of profitable services
Per diem	One day	Shared	Longer stays, with less done each day
Per case, such as a DRG	One admission	Provider, within the stay	Shorter stays, more admissions, upcoding, cost shifted after discharge
Bundle or episode	An episode across settings	Provider	Avoiding complex patients, more episodes started
Capitation	One person per month	Provider	Under-provision and enrollment of healthier people
Salary	One clinician per year	Employer	Low productivity absent other management
Global budget	An institution per year	Provider	Queues and restricted access at the margin

Read the last column downward and you can see the whole history of payment reform. Every reform in the past forty years has been a move down this table, away from paying for volume. Every one has then met the distortion in its own row, and the next reform has been an attempt to patch it.

How a fee-for-service price is actually set

People say fee-for-service as though it means a market price. In Medicare it means something quite specific and quite administered.

Since 1992 Medicare has paid physicians using a resource-based relative value scale, developed by a team led by the economist William Hsiao at Harvard. Each service is assigned relative value units in three components: physician work, capturing time, skill, and intensity; practice expense, covering staff, rent, and supplies; and professional liability insurance. Those units are adjusted for local input costs and multiplied by a conversion factor, a single dollar amount set nationally. Payment equals adjusted relative value units times the conversion factor. Change the conversion factor and every physician price in the country moves together.

The relative values themselves are recommended to Medicare by a committee convened by the American Medical Association, composed largely of specialty society representatives. Medicare has historically accepted the great majority of its recommendations. The persistent criticism is structural rather than personal: a committee dominated by proceduralists, valuing procedures against office visits, has produced a schedule under which procedural specialties earn far more per hour than primary care. That relative valuation then propagates far beyond Medicare, because commercial insurers commonly set their own fees as a percentage of Medicare rates.

For hospitals the equivalent machinery is the diagnosis-related group. Patients are classified by principal diagnosis, procedures, complications, and comorbidities into groups expected to consume similar resources. Each group carries a weight; payment is the weight times a base rate adjusted for local wages, teaching status, and the share of low-income patients served, with extra outlier payments for extreme cases. The seam here is documentation: the more thoroughly complications and comorbidities are recorded, the higher the weight. Recording them more thoroughly is both better medicine and more revenue, which makes DRG creep genuinely hard to distinguish from improved coding.

Why this matters: Medicare prices are administered, not discovered, and the relative values behind them are recommended by a committee whose composition shapes what American medicine pays best.

Capitation, and why it retreated

Capitation pays a fixed amount per enrolled person per month, whatever they use. It inverts every fee-for-service incentive at a stroke: prevention becomes profitable, an avoided admission is money kept, and there is no reward for volume.

It also creates two problems that killed its first American expansion. The first is stinting: if care costs the provider money, the cheapest patient is one who receives nothing. The second is selection: a capitated payment that does not adjust for health status makes healthy enrollees profitable and sick ones ruinous, which rewards marketing to the well.

Both showed up in the managed care expansion of the 1990s, along with utilization review procedures that patients and physicians experienced as arbitrary. The backlash was fierce, produced state patient protection laws, and ended the growth of tightly capitated arrangements outside a few integrated systems. The technical fix for selection, risk adjustment, has since improved considerably, which is why capitation has returned in the form of Medicare Advantage and of accountable care contracts. The fix for stinting is quality measurement, which is the subject of Lesson 9 and is harder than it looks.

What the value-based experiments actually found

The ACA created the Center for Medicare and Medicaid Innovation with authority to test payment models and expand those that work. It has run dozens. Here is what the serious evaluations say, and the honest summary is more modest than the rhetoric on either side.

• Accountable care organizations. Groups of providers accept accountability for the total cost and quality of a defined population, keeping a share of savings against a benchmark. Evaluations of the Medicare Shared Savings Program find real but small net savings, and a consistent split: ACOs led by physician groups saved meaningfully, while hospital-integrated ACOs saved little or nothing. The explanation is that a physician group profits from keeping patients out of a hospital it does not own, while a hospital system loses admissions it would have been paid for.
• Bundled payment. The mandatory joint replacement bundle produced episode spending roughly 3 percent lower than in control markets, achieved almost entirely by sending fewer patients to inpatient rehabilitation and skilled nursing, with no detected increase in complications. That is a genuine, if narrow, success, and it worked because joint replacement is a well defined, largely elective episode.
• Readmission penalties. The Hospital Readmissions Reduction Program penalized hospitals with high 30-day readmission rates for selected conditions from 2013. Readmissions fell. Then the arguments began: some of the fall reflects patients being placed in observation status rather than admitted, and one prominent analysis found increased 30-day mortality for heart failure and pneumonia patients over the same period, which other researchers dispute on methodological grounds. It remains the most contested evaluation in the field.
• Hospital value-based purchasing. Redistributing a small share of Medicare payment according to quality scores has produced little detectable improvement in the outcomes it measures.
• Physician quality programs. The Merit-based Incentive Payment System, created by the 2015 law that repealed the sustainable growth rate formula, imposes substantial reporting burden, and studies find its scores correlate weakly with the outcomes patients care about.

Then the summary judgment. In 2023 the Congressional Budget Office assessed the Innovation Center's first decade and concluded that, taken together, its models had increased net federal spending rather than reducing it, by several billion dollars, once the cost of running them and the shared savings paid out were counted. A handful of models saved money; more did not; and the statutory authority to expand successful ones has been used sparingly.

What should you conclude? Not that payment reform is futile. The bundle result and the physician-group ACO result are real. The conclusion is narrower and more useful: models work when the unit of payment matches a well defined clinical episode and when the entity bearing risk actually profits from the savings. They fail when the risk-bearing entity owns the capacity that the savings would leave empty.

Bottom line: After a decade of experiments the Innovation Center's portfolio raised net federal spending, while the specific models that worked shared two features: a clearly bounded episode, and a risk-bearer with nothing to lose from reduced volume.

Common misconceptions

• Fee-for-service means prices set by a market. In Medicare, and therefore indirectly in much of the commercial market, they are administered prices built from relative value units and a conversion factor.
• Capitation means denying care. It means the provider keeps what is not spent, which rewards prevention as strongly as it tempts stinting. Which effect dominates depends on risk adjustment and quality measurement.
• Value-based payment has replaced fee-for-service. Most American health care is still paid for by the service. Alternative models usually sit on top of a fee-for-service chassis, settling up after the fact against a benchmark.
• Upcoding is straightforward fraud. Documenting a comorbidity that is genuinely present is both good practice and more revenue, which is exactly why distinguishing better coding from gaming is so difficult.
• The readmissions program simply worked. Readmissions fell, but observation stays rose and the mortality evidence is contested, which is why it is the standard cautionary tale about single-metric incentives.

Putting it together

• Medicare's prospective payment system, effective October 1983, converted hospital days from revenue into cost and shortened stays quickly.
• Payment methods can be ordered by unit size; as the unit grows, risk shifts to the provider and the distortion flips from too much care to too little.
• Physician payment rests on relative value units for work, practice expense, and liability, times a national conversion factor, with relative values recommended by an AMA-convened committee.
• Diagnosis-related groups pay per admission by weight, creating a documentation seam that makes coding intensity hard to separate from better recording.
• Capitation reverses volume incentives and introduces stinting and selection, addressed respectively by quality measurement and risk adjustment.
• Physician-group ACOs saved money and hospital-integrated ones largely did not; the mandatory joint replacement bundle cut episode spending about 3 percent through less post-acute care.
• CBO found the Innovation Center's first decade increased net federal spending, which narrows rather than refutes the case for payment reform.

Sources

1. McWilliams, J. M., Hatfield, L. A., Landon, B. E., Hamed, P., and Chernew, M. E. (2018). Medicare spending after 3 years of the Medicare Shared Savings Program. New England Journal of Medicine, 379(12), 1139-1149. doi.org
2. Barnett, M. L., Wilcock, A., McWilliams, J. M., Epstein, A. M., Joynt Maddox, K. E., Orav, E. J., Grabowski, D. C., and Mehrotra, A. (2019). Two-year evaluation of mandatory bundled payments for joint replacement. New England Journal of Medicine, 380(3), 252-262. doi.org
3. Wadhera, R. K., Joynt Maddox, K. E., Wasfy, J. H., Haneuse, S., Shen, C., and Yeh, R. W. (2018). Association of the Hospital Readmissions Reduction Program with mortality among Medicare beneficiaries. JAMA, 320(24), 2542-2552. doi.org
4. Center for Medicare and Medicaid Innovation. (n.d.). Innovation Models. innovation.cms.gov
5. Congressional Budget Office. (n.d.). Publications. cbo.gov

A 1984 bargain that still governs the market

The Drug Price Competition and Patent Term Restoration Act of 1984, universally called Hatch-Waxman after its sponsors, is the reason that nearly nine of every ten prescriptions filled in the United States today are for generic drugs. It was a trade. Generic manufacturers got an abbreviated approval pathway: instead of repeating the original clinical trials, a generic maker need only demonstrate that its product is bioequivalent to the reference drug, meaning it delivers the same active ingredient to the bloodstream at a comparable rate and extent. Brand manufacturers got compensation for the patent life consumed by regulatory review, plus periods of marketing exclusivity independent of patents.

The bargain worked, and it worked lopsidedly. Generic drugs are close to nine tenths of American prescriptions and well under a fifth of American drug spending. The country has, by international standards, an extremely cheap generic market and an extremely expensive branded one. Any serious account of drug prices has to explain both halves, and most popular accounts explain neither.

Key idea: The United States has two drug markets: a fiercely competitive generic market that produces some of the lowest prices in the world, and a branded market with the highest prices in the world. Hatch-Waxman created the first and constrained the second only at the margins.

How a drug reaches the market

The Food and Drug Administration's authority rests on statutes passed after disasters. The 1938 Food, Drug, and Cosmetic Act, requiring proof of safety before marketing, followed the elixir sulfanilamide poisonings of 1937. The 1962 Kefauver-Harris Amendments, requiring proof of effectiveness as well, followed the thalidomide episode. That pattern, regulation arriving after harm, is worth noting because it recurs throughout this course.

Stage	What happens	Typical scale
Preclinical	Laboratory and animal testing before any human exposure	Years; most candidates fail here
Investigational New Drug application	FDA reviews the plan for human testing	30 days for the agency to object
Phase 1	Safety and dose finding, usually in healthy volunteers	Tens of participants
Phase 2	Preliminary efficacy and further safety in patients	Hundreds
Phase 3	Confirmatory efficacy against control	Hundreds to thousands
New Drug or Biologics License Application	Full review of the evidence, manufacturing, and labeling	Months, funded partly by industry user fees since 1992
Phase 4	Post-marketing surveillance and required confirmatory studies	Ongoing

Two pathways complicate that picture and both are worth knowing. Accelerated approval, created in 1992 under pressure from AIDS activists, permits approval on a surrogate endpoint, a laboratory measure reasonably likely to predict clinical benefit, with confirmatory trials required afterward. It has plainly saved lives in oncology and HIV. It has also produced approvals where the confirmatory evidence never arrived or arrived negative, and for years the agency had no practical mechanism to withdraw them, a gap Congress addressed in 2022.

The clearest case is the Alzheimer's drug aducanumab, approved in 2021 on the surrogate endpoint of reduced amyloid plaque after the agency's own advisory committee voted overwhelmingly against approval. Three committee members resigned. The launch price was set near 56,000 dollars a year and then cut by half after almost nobody bought it, and Medicare ultimately restricted coverage to patients in trials. The episode is the standard teaching case for the tension between speed and certainty, and neither side of that tension is frivolous: a patient dying of a disease with no treatment is harmed by delay just as surely as by a drug that does not work.

The Orphan Drug Act of 1983 pulls in a different direction. It offers seven years of market exclusivity plus tax credits for drugs treating conditions affecting fewer than 200,000 Americans, and it demonstrably worked: rare disease treatments went from a trickle to hundreds of approvals. The criticism is that the incentives can be captured by dividing a common disease into narrow subtypes, each qualifying separately.

What matters here: Approval standards are a policy dial, not a scientific constant. Loosening them buys speed for desperate patients and pays for it in approvals that later prove worthless, and the aducanumab case is what that trade looks like in practice.

Following one prescription dollar

Now the part that confuses everyone, including most people who work in the industry. American drug prices come in at least four flavors, and quoting the wrong one is the most common error in public argument.

Price	What it means	Who actually faces it
List price	The manufacturer's published wholesale price	Uninsured patients, and insured patients paying coinsurance or in a deductible
Net price	List minus rebates and discounts	The plan or insurer
Negotiated or contracted price	What a pharmacy is reimbursed	The pharmacy
Statutory price	Prices set by law, including Medicaid rebates, 340B discounts, and federal supply schedule prices	Medicaid, safety-net hospitals, the VA

Between the manufacturer and the patient sits the pharmacy benefit manager, an entity that designs formularies, negotiates rebates, and processes claims for health plans. Three PBMs handle roughly four fifths of American prescriptions, and each is now owned by or affiliated with a large insurer.

Here is the mechanism that makes drug pricing so strange. A manufacturer wanting favorable formulary placement offers a rebate off list. A higher list price allows a larger rebate, which makes the offer more attractive, so both list prices and rebates rise together while the net price the plan pays grows slowly. Analysts call the widening gap the gross-to-net bubble. The patient with a deductible or percentage coinsurance pays a share of the list price, not the net price, so the person least able to bear it faces the number that has been inflated on purpose.

That is why the reforms in this area cluster around transparency and around delinking PBM compensation from list price. It is also why any policy that caps a list price without touching the rebate machinery can leave net prices, and therefore premiums, roughly where they were.

So what?: List and net prices move apart because rebates are calculated off list, and the patient in a deductible pays off list, which is the single cleanest explanation of why American drug costs feel so much worse than the aggregate spending data suggests.

Why the United States pays more

RAND researchers, comparing American prices with those in dozens of other wealthy countries, found overall American prices roughly two and a half times higher, and for brand-name drugs roughly three and a half times higher. The generic comparison runs the other way: American generics are cheaper than most.

The explanation is institutional rather than mysterious. Other wealthy countries have a single national body that either negotiates or sets prices, usually with an explicit assessment of whether the drug is worth its price, and with the credible option of not covering it. The United States has no such body. Medicare, the largest purchaser, was expressly forbidden by the 2003 law creating Part D from negotiating drug prices, a provision known as the noninterference clause. Private plans negotiate individually and are legally constrained in which drugs they may exclude, particularly in protected classes.

The Inflation Reduction Act of 2022 changed part of this. It required the Secretary of Health and Human Services to negotiate maximum fair prices for a growing list of high-spending drugs, beginning with ten Part D drugs selected in 2023 whose negotiated prices take effect in 2026, with additional drugs selected annually and Part B drugs joining later. It imposed rebates when a drug's price rises faster than inflation. It capped insulin cost sharing at 35 dollars a month for Medicare beneficiaries and capped annual Part D out-of-pocket spending at 2,000 dollars from 2025, replacing a catastrophic tier that had no ceiling at all. The law applies only to Medicare, and manufacturers have challenged it in court, so treat the current state of play as something to check rather than to memorize.

What research and development actually costs

The industry's central argument is that high American prices fund the research that produces new medicines, and that the rest of the world free-rides on it. Take that argument seriously and then look at the numbers, because they are genuinely contested.

The most quoted estimate comes from the Tufts Center for the Study of Drug Development, which put the capitalized cost per approved drug at about 2.6 billion dollars. That figure includes the cost of failures and, importantly, includes the opportunity cost of capital over a decade of development, which accounts for roughly half of it. Critics note that the underlying company data are confidential and unauditable.

A second team, working from public filings for companies whose only approvals came in a defined window, estimated a median of around one billion dollars per approved drug, with wide variation by therapeutic area. The two estimates are not measuring quite the same thing, and neither is obviously dishonest, but the gap is large enough that any policy argument resting on a specific number should be treated cautiously.

The other half of the picture is public investment. Researchers examining every new drug approved by the FDA from 2010 to 2016 found that National Institutes of Health funding had contributed to published research associated with all 210 of them, with the public contribution concentrated in the basic science that identifies a target. That does not make private development costless, and translating a target into an approved medicine is expensive and mostly unsuccessful. It does mean the story of a drug's origin is usually a public discovery followed by private development, and pricing arguments that ignore either half are incomplete.

The upshot: Estimates of the cost of bringing a drug to market differ by more than a factor of two depending on method, and public funding contributed to the underlying science of every drug approved over a seven-year window, so neither the pure innovation defense nor the pure profiteering charge survives contact with the evidence.

Common misconceptions

• Drug spending is the main driver of American health costs. Retail prescription drugs are roughly a tenth of national health expenditure. They are salient because patients pay for them directly and repeatedly.
• Generic drugs are inferior copies. A generic must demonstrate bioequivalence and meet the same manufacturing standards. American generics are among the cheapest in the world and constitute the overwhelming majority of prescriptions.
• The list price is what someone pays. Plans pay net of rebates. The people most likely to face the list price are the uninsured and patients in deductibles, which inverts the usual logic of insurance.
• FDA approval means a drug works better than existing options. Approval requires evidence of efficacy against the standard used in the trial, often placebo. Comparative effectiveness against existing treatments is frequently unknown at launch.
• Medicare has always negotiated its drug prices. The 2003 law creating Part D expressly forbade it, and negotiation for a limited set of drugs began only under the 2022 law.
• High prices are simply necessary to fund research. Research is genuinely expensive and its cost is disputed by a factor of more than two, and the basic science behind approved drugs is heavily publicly funded.

What to carry forward

• Hatch-Waxman in 1984 created the abbreviated generic pathway in exchange for patent restoration and exclusivity, producing a cheap generic market and an expensive branded one.
• Approval runs from investigational new drug through three trial phases to a new drug or biologics licence, with accelerated approval permitting surrogate endpoints and confirmatory trials afterward.
• The aducanumab approval, over the advisory committee's objection and at a launch price near 56,000 dollars, is the standard case study in the speed versus certainty trade.
• List, net, contracted, and statutory prices are four different numbers, and rebates calculated off list push list prices upward.
• Three pharmacy benefit managers handle roughly four fifths of American prescriptions.
• American brand prices run roughly three and a half times those in comparison countries, chiefly because no single body negotiates or sets them.
• The Inflation Reduction Act introduced Medicare negotiation, inflation rebates, a 35 dollar insulin cap, and a 2,000 dollar Part D out-of-pocket ceiling.
• Development cost estimates range from around one billion to about 2.6 billion dollars per approved drug, and public funding contributed to the science behind all 210 drugs approved from 2010 to 2016.

Sources

1. United States Food and Drug Administration. (n.d.). Drugs. fda.gov
2. Cleary, E. G., Beierlein, J. M., Khanuja, N. S., McNamee, L. M., and Ledley, F. D. (2018). Contribution of NIH funding to new drug approvals 2010-2016. Proceedings of the National Academy of Sciences, 115(10), 2329-2334. doi.org
3. Wouters, O. J., McKee, M., and Luyten, J. (2020). Estimated research and development investment needed to bring a new medicine to market, 2009-2018. JAMA, 323(9), 844-853. doi.org
4. RAND Corporation. (n.d.). Prescription Drug Prices. rand.org
5. KFF. (n.d.). Prescription Drugs. kff.org

The surgeons who stopped operating on the sickest patients

New York State began collecting risk-adjusted mortality data for coronary artery bypass surgery in 1989. After a newspaper won a freedom of information lawsuit, surgeon-level results became public in the early 1990s. Pennsylvania followed. This was, on its face, exactly what patients should want: an outcome that matters, adjusted for how sick the patients were, published so anyone could look.

Mortality rates in both states fell. Then economists went looking at what else had changed. David Dranove, Daniel Kessler, Mark McClellan, and Mark Satterthwaite compared Medicare patients in report-card states with those elsewhere and found that surgeons had shifted toward healthier patients. Sicker patients received surgery less often, or received it later, or were transferred out of state. For the healthiest patients, the matching between patient and high-quality surgeon improved. For the sickest, outcomes got worse. The authors' overall assessment was that the report cards, taken as a whole, reduced patient welfare.

Nobody committed fraud. The surgeons did not falsify records. They declined to operate on patients whose deaths would appear on their published record, which is precisely the behavior an outcome-based scorecard rewards. That is the central problem of this lesson and it has no clean solution, only better and worse ways of managing it.

Key idea: A published outcome measure changes which patients get treated, not only how well they are treated, and the effect falls hardest on the patients whose cases are hardest.

Donabedian's three categories, and why they still organize the field

In 1966 Avedis Donabedian proposed that the quality of medical care can be assessed in three ways, and sixty years later every quality program in the world is still built from his categories.

Category	What it measures	Example	Strength and weakness
Structure	The stable characteristics of the setting	Nurse-to-patient ratio, accreditation, presence of an intensivist	Easy to verify and slow to change; only loosely tied to what happens to a patient
Process	What was actually done	Aspirin given on arrival for heart attack, screening completed on schedule	Directly actionable and attributable; only as good as the evidence behind the process
Outcome	What happened to the patient	30-day mortality, surgical site infection, functional recovery	What everyone cares about, and the hardest to attribute fairly
Experience	What care was like for the patient	Standardized patient survey scores on communication and responsiveness	Captures what only the patient can know; confounded by amenities and expectations

Two national reports set the modern agenda. The Institute of Medicine's To Err Is Human, published in 1999, estimated that between 44,000 and 98,000 Americans died each year from preventable medical errors, and reframed those deaths as failures of systems rather than of individuals. Its successor, Crossing the Quality Chasm, published in 2001, defined six aims: care should be safe, effective, patient-centered, timely, efficient, and equitable. Those six are still the organizing scheme for most quality programs.

A later paper claimed medical error was the third leading cause of death in the United States at around 250,000 deaths a year. Treat that figure carefully. It extrapolated from a small number of studies with limited samples and did not establish that the patients would have survived absent the error. The underlying point, that preventable harm is common, does not need the inflated number, and using it makes the argument easier to dismiss.

Three problems that limit every quality score

Before you read any public quality rating, know these three.

Risk adjustment is never complete. A hospital treating sicker patients will have worse raw outcomes. Statistical adjustment uses the variables that happen to be recorded, typically diagnoses, age, and a few clinical values. It cannot adjust for frailty, social support, housing, or the reason a family chose this hospital. Under-adjustment penalizes safety-net hospitals for their patients. Over-adjustment, using variables that are themselves consequences of poor care, hides real differences. There is no setting of the dial that is right for both.

Attribution is contestable. A patient with heart failure sees a cardiologist, a primary care physician, a home health nurse, and two hospitals in a year. Whose outcome is it? Programs assign the patient to whoever provided the plurality of care, or to the discharging hospital, and every rule creates a set of providers held responsible for decisions they did not make.

Small numbers destroy reliability. If a hospital performs 40 of a given operation a year with an expected mortality of 2 percent, its expected number of deaths is under one. One extra death doubles its rate. Rankings built on such counts are largely reordering noise, which is why careful analysts use funnel plots that widen the acceptable range for low-volume providers rather than publishing a league table.

Remember: Incomplete risk adjustment, contested attribution, and small denominators mean that most differences in a published quality ranking are not differences in quality.

Campbell's law in a hospital

The social scientist Donald Campbell put the general principle plainly in the 1970s: the more a quantitative indicator is used for social decision-making, the more it will distort and corrupt the processes it was meant to monitor. Charles Goodhart made the same point about economic statistics. Health care supplies unusually well documented examples.

• Cardiac report cards. Publishing surgeon-level mortality shifted surgery away from the sickest patients, as above.
• Emergency waiting time targets. Britain's requirement that patients leave the emergency department within four hours produced real reductions in waiting, and also produced clock management: patients held in ambulances outside the door, decisions to admit made near the deadline, and a visible spike in departures in the minutes before four hours elapsed.
• Readmission penalties. As Lesson 7 described, some of the reduction in readmissions was a reclassification into observation status, which does not count as an admission.
• Sepsis bundles. A measure requiring broad-spectrum antibiotics within a fixed window of suspected sepsis saves lives when sepsis is present and drives antibiotic use in patients who do not have it, with consequences for resistance that the measure does not count.
• Pay for performance in general practice. Britain's Quality and Outcomes Framework attached a large share of practice income to dozens of indicators from 2004. Measured performance improved, though often continuing trends already underway, and when incentives were later removed from a set of indicators, performance on them declined.

The pattern across all five is the same. The measure captures part of the goal. The part it captures improves. The part it omits moves, and the direction of that movement depends on what was easiest to do. That is not a reason to stop measuring; unmeasured care was not better, it was merely unexamined. It is a reason to pair every incentive with a balancing measure that would detect the predictable evasion.

Why this matters: The right response to gaming is not to abandon measurement but to name the evasion in advance and measure that too, which is a discipline most programs skip.

When measurement worked

An honest lesson has to include the other column, and it is a strong one.

Central line-associated bloodstream infections in intensive care were long treated as an unavoidable cost of critical care. In 2003 a project across more than 100 intensive care units in Michigan combined a five-item evidence-based checklist for line insertion with unit-level infection tracking, an explicit programme to change the safety culture, and permission for a nurse to stop a physician who skipped a step. Infection rates fell to near zero and stayed there for the study period. That result has been reproduced widely and represents one of the clearest quality successes on record.

Notice what made it work. The outcome was well defined and countable. The process linked to it was known and cheap. The measurement was local and immediate rather than a public ranking. And the intervention changed the authority structure of the room, not just the paperwork.

Compare a surgical safety checklist. An eight-hospital international study published in 2009 found substantial reductions in death and complications after introducing a nineteen-item checklist. When Ontario later mandated checklist use across the province, a subsequent analysis found no significant improvement in mortality or complications. Same tool, different result. The most plausible explanation is that the first study measured genuine implementation, with training and buy-in, while the mandate measured compliance with a form. That distinction, between doing the thing and recording that you did the thing, is the whole ball game in quality improvement.

The cost of measuring

Measurement is not free, and the bill is larger than most people assume. A study of American physician practices estimated that they spent, on average, more than 780 hours per physician per year dealing with quality measure reporting, at a national cost above 15 billion dollars a year. Hospitals report to Medicare, to state agencies, to accreditors, to multiple commercial payers, and to specialty registries, each with its own definitions of ostensibly the same measure.

That burden has two costs. The obvious one is money and clinician time. The subtler one is that when a clinician spends an hour on documentation for a measure, the hour comes from somewhere, and there is decent evidence that documentation burden is a major contributor to clinician burnout, which has its own effects on care. Any argument for a new measure has to clear that bar, and most proposals never attempt it.

Common misconceptions

• Outcome measures are always better than process measures. Outcomes are what matters and are the hardest to attribute; process measures are directly actionable and fairly attributable. Good programs use both.
• Risk adjustment makes comparisons fair. It makes them fairer, using the variables that happen to be recorded. Frailty, housing, and social support are rarely among them.
• A hospital with a five-star rating is better than one with three stars. With small denominators and different rating methodologies, much of the difference is noise, and different rating systems routinely disagree about the same hospital.
• Gaming means providers are dishonest. Most documented distortions involve entirely legal, clinically defensible choices that happen to improve the score, which is why they are so hard to prevent.
• Checklists work. Checklists work when they are implemented with training, buy-in, and a change in who is permitted to stop the line. Mandated as paperwork, they have not reproduced their original results.
• Medical error is the third leading cause of death. That widely repeated figure rests on extrapolation from small samples and does not establish that the patients would otherwise have survived.

Where this leaves us

• Donabedian's structure, process, and outcome categories, with patient experience added, still organize every quality program.
• To Err Is Human in 1999 reframed error as a systems problem; Crossing the Quality Chasm in 2001 set six aims: safe, effective, patient-centered, timely, efficient, and equitable.
• Every quality score is limited by incomplete risk adjustment, contested attribution, and unreliable small denominators.
• New York and Pennsylvania cardiac report cards shifted surgery away from the sickest patients, worsening their outcomes.
• Emergency waiting targets, readmission penalties, sepsis bundles, and general practice pay-for-performance all show the measured part improving and the unmeasured part moving.
• The Michigan intensive care line-infection project worked because the outcome was countable, the process was known, feedback was local, and the authority structure changed.
• A surgical checklist that reduced deaths in a study of eight hospitals produced no measurable benefit when mandated province-wide, which is the difference between implementation and compliance.
• Quality reporting costs American physician practices more than 780 hours per physician per year and over 15 billion dollars annually.

Sources

1. Dranove, D., Kessler, D., McClellan, M., and Satterthwaite, M. (2003). Is more information better? The effects of report cards on health care providers. Journal of Political Economy, 111(3), 555-588. doi.org
2. Pronovost, P., Needham, D., Berenholtz, S., Sinopoli, D., Chu, H., Cosgrove, S., et al. (2006). An intervention to decrease catheter-related bloodstream infections in the ICU. New England Journal of Medicine, 355(26), 2725-2732. doi.org
3. Casalino, L. P., Gans, D., Weber, R., Cea, M., Tuchovsky, A., Bishop, T. F., Miranda, Y., Frankel, B. A., Ziehler, K. B., Wong, M. M., and Evenson, T. B. (2016). US physician practices spend more than 15.4 billion dollars annually to report quality measures. Health Affairs, 35(3), 401-406. doi.org
4. Agency for Healthcare Research and Quality. (n.d.). Quality and Patient Safety. ahrq.gov
5. Donabedian, A. (1966). Evaluating the quality of medical care. Milbank Memorial Fund Quarterly, 44(3), 166-206.

The paper that settled an argument

In 2003 Gerard Anderson, Uwe Reinhardt, Peter Hussey, and Varduhi Petrosyan published a paper in Health Affairs with a title that was itself the finding: It's the Prices, Stupid. They compared the United States with other wealthy countries and asked the obvious question. If Americans spend far more, do they get more?

They did not. Americans had fewer physician visits per person than the average wealthy country. They were admitted to hospital no more often. They had fewer practising physicians per capita and fewer hospital beds. On most measures of how much medical care people consumed, the United States was ordinary or below average. What was not ordinary was the price of each unit: a day in an American hospital, an American physician's fee, an American branded drug, all cost multiples of their equivalents elsewhere.

Twenty years of subsequent work has complicated the picture in details, particularly around imaging, specialist care, and administrative complexity, and has confirmed its core. This lesson takes the 4.9 trillion dollars from Lesson 1 and takes it apart: what it buys, who pays it, why it is so large, and what fraction of it is doing no good.

Key idea: The United States does not consume more medical care than other wealthy countries. It pays much more per unit, which means most cost-control ideas aimed at volume are aimed at the wrong variable.

The money by category

The Centers for Medicare and Medicaid Services publish national health expenditure accounts every year, and reading them once carefully is worth more than reading a hundred opinion pieces. Approximate shares of the 2023 total look like this.

Category	Approximate share	Note
Hospital care	Around 30 percent	The single largest category by a wide margin
Physician and clinical services	Around 20 percent	Includes clinic-based services billed by physician groups
Retail prescription drugs	Around 9 percent	Excludes drugs administered in hospitals and clinics, which sit in those categories
Nursing homes and home health	Around 8 percent	The long-term care Medicare largely does not cover
Dental and other professional services	Around 7 percent	Mostly outside the main insurance system
Net cost of private insurance and government administration	Around 8 percent	Administration as counted in the accounts, which understates the total
Government public health activity	Low single digits	The entire prevention apparatus, discussed in Lesson 12
Investment in research and structures	Around 5 percent	Buildings, equipment, and research

Three things in that table repay attention. Hospital care plus physician services is half of everything, which is why any serious cost strategy has to engage with hospitals and physicians rather than with the categories that are easier to attack politically. Retail drugs are about a tenth, far less than their share of public argument, though the figure understates the total because clinic-administered drugs are counted elsewhere. And government public health activity, the entire apparatus of disease surveillance, inspection, and prevention, is a rounding error next to treatment.

Now the same money by source. Roughly speaking, private health insurance pays about 30 percent, Medicare about 21 percent, Medicaid about 18 percent, and households directly out of pocket about 11 percent, with the remainder from other public programs, other private revenue, and investment. Put the public programs together and add the tax exclusion from Lesson 3, and government is financing considerably more than half of American health care, which is a fact that both political parties tend to avoid saying plainly.

Price times quantity, and which one is the problem

Spending equals price times quantity. It sounds trivial. It is the single most clarifying identity in this course, because almost every proposed solution addresses one term or the other, and confusing them wastes decades.

The evidence on the quantity side is consistent. Compared with other wealthy countries, the United States has fewer physicians per capita, fewer hospital beds per capita, and fewer physician visits and hospital discharges per person. It does use more of certain things: imaging, particularly magnetic resonance and computed tomography, and some high-margin procedures.

The evidence on the price side is stark. Studies of employer claims data have found commercial payers paying well over twice what Medicare pays for the same hospital services, with variation across states and systems that dwarfs any plausible difference in quality. Brand-name drug prices run several times those in peer countries. A physician in the United States earns considerably more than one in France or Germany. And prices vary enormously within the United States too: analyses of commercial claims routinely find the same imaging study priced at ten times as much a few miles away.

That last point deserves emphasis because it separates two different problems. Geographic variation in Medicare spending, which the Dartmouth Atlas researchers documented for decades, is mostly variation in quantity, since Medicare's prices are administratively set. Geographic variation in commercial spending is mostly variation in price, since utilization is more uniform than the spending. The National Academies examined this in 2013 and concluded, importantly, that variation does not track neatly to geography in a way that would let a payer reward efficient regions, because the variation within regions is as large as the variation between them.

The point: Public program spending varies mainly because of how much care is delivered; commercial spending varies mainly because of what is charged, and the two require entirely different remedies.

Administration, the category everyone argues about

The national accounts put the net cost of private insurance and government administration at roughly 8 percent. That number counts insurer overhead and government program administration, and it does not count what providers spend responding to insurers: the billing staff, the prior authorization phone calls, the coding specialists, the denial appeals, the eligibility checks.

Adding those in raises the estimate substantially, and by how much is contested. Comparisons of the United States with Canada, whose single-payer provinces have far simpler billing, have estimated American administrative spending at around a third of total health expenditure, against roughly half that in Canada. Other researchers put the American figure lower, arguing that some activities counted as administration are genuinely care coordination. Micro-studies that follow specific encounters have measured billing and insurance-related costs ranging from a few percent of revenue for a simple primary care visit to a quarter for some inpatient procedures.

What is not disputed is the direction and its cause. A country with one payer, one form, and one price schedule spends less on administration than a country where every hospital negotiates separately with dozens of insurers, each with its own formulary, prior authorization rules, network, and appeals process. The administrative burden is the price of the pluralism, and it is paid on both sides of every transaction.

Who spends the money

Recall the skew from Lesson 2 and now put national numbers on it. In the household survey data, the top 5 percent of spenders account for roughly half of all personal health spending in a year, and the top 1 percent for about a fifth. The bottom half of the population accounts for a few percent.

Two policy implications follow, and they pull against each other. The first is that any strategy aiming at the average patient is aiming at almost none of the money, which is the case for care management programs targeted at high-cost patients. The second is a caution: high spenders in one year are frequently not high spenders the next, because much high spending follows an acute event, a cancer diagnosis, a hip fracture, a premature birth. Programs that identify last year's expensive patients and manage them intensively often show savings that are mostly regression to the mean. Evaluations of care management programs have repeatedly found this, which is why the credible ones use randomized designs.

Worth holding on to: Spending is extremely concentrated and only partly persistent, so targeting last year's high spenders produces apparent savings that a randomized comparison usually removes.

Why it grows

Over the long run American health spending has grown faster than the economy, and the standard decomposition attributes that excess growth to a few sources.

• Technology. Joseph Newhouse argued in 1992 that the expansion of what medicine can do accounts for the largest share of long-run growth, and the argument has held up. New capability is mostly additive: intensive care, transplantation, biologics, and imaging did not replace older care so much as add to it.
• Prices. Discussed above, and the dominant explanation for the American level rather than for the universal upward trend.
• Income. Richer societies spend more on health, and the relationship is strong across countries.
• Aging. Real but far smaller than intuition suggests. Demographic change accounts for a modest share of growth, because the cost of care in the final years of life has changed less than the cost of care at every age.
• Insurance coverage itself. Expanding coverage raises use, as Lesson 2 established.

Then the waste question. Two prominent estimates, one in 2012 and one in 2019, put American health care waste at roughly a quarter of total spending, dividing it into failures of care delivery, failures of care coordination, overtreatment, administrative complexity, pricing failures, and fraud and abuse. Take those numbers as an order of magnitude rather than a target. The 2019 estimate found that the interventions with demonstrated savings recovered only a small fraction of the estimated waste, which is the crucial and least-quoted part of the paper. Identifying waste in aggregate is easy. Removing a specific dollar of it, prospectively, without also removing care someone needed, is the problem nobody has solved.

Common misconceptions

• Americans overuse medical care. On most measures of volume they are average or below. What is high is the price per unit.
• Drug spending drives the cost problem. Retail drugs are around a tenth of the total. Hospital and physician services together are about half.
• An aging population explains rising costs. Demography contributes modestly; expanding technological capability contributes far more.
• Health care in the United States is mostly private. Once Medicare, Medicaid, other public programs, public employee coverage, and the tax exclusion are counted, government finances well over half of it.
• Identifying a quarter of spending as waste means it could be cut. The same research finds that interventions with demonstrated savings recover only a small share of it, because waste is far easier to see in aggregate than to remove case by case.
• Managing high-cost patients reliably saves money. Much high spending is transient, so apparent savings from targeting last year's high spenders often reflect regression to the mean rather than program effect.

The takeaway

• Hospital care is roughly 30 percent of national health spending and physician and clinical services roughly 20 percent, with retail drugs around 9 percent.
• By source, private insurance pays around 30 percent, Medicare around 21 percent, Medicaid around 18 percent, and households out of pocket around 11 percent.
• Anderson and colleagues showed in 2003 that American utilization is ordinary while American prices are not, and later work has confirmed it.
• Medicare spending varies geographically mainly by quantity; commercial spending varies mainly by price, so the two need different remedies.
• Administrative spending is roughly 8 percent in the official accounts and far higher once provider-side billing costs are counted, with estimates up to about a third of total spending.
• The top 5 percent of spenders account for about half of spending and the top 1 percent for about a fifth, but high spending is only partly persistent from year to year.
• Technology is the largest long-run driver of growth; aging contributes far less than commonly assumed.
• Waste is estimated at about a quarter of spending, and the interventions with proven savings recover only a small fraction of it.

Sources

1. Anderson, G. F., Reinhardt, U. E., Hussey, P. S., and Petrosyan, V. (2003). It's the prices, stupid: Why the United States is so different from other countries. Health Affairs, 22(3), 89-105. doi.org
2. Shrank, W. H., Rogstad, T. L., and Parekh, N. (2019). Waste in the US health care system: Estimated costs and potential for savings. JAMA, 322(15), 1501-1509. doi.org
3. Centers for Medicare and Medicaid Services. (n.d.). National Health Expenditure Data. cms.gov
4. Agency for Healthcare Research and Quality. (n.d.). Medical Expenditure Panel Survey. meps.ahrq.gov
5. Newhouse, J. P. (1992). Medical care costs: How much welfare loss? Journal of Economic Perspectives, 6(3), 3-21.

A chancellor who invented health insurance to defeat socialists

On 15 June 1883 the German Reichstag passed the Krankenversicherungsgesetz, the Health Insurance Act, requiring industrial workers to join a sickness fund with contributions shared between worker and employer. It is the oldest national health insurance law in the world, and Otto von Bismarck, who drove it through, was not a social reformer. He had banned the Social Democratic Party's organizations five years earlier and was looking for a way to detach workers from the socialist movement by having the state meet their material needs first. He said so.

That origin is worth knowing for two reasons. It shows that universal coverage is not the property of any political tradition; it was invented by a conservative chancellor as a counter-revolutionary measure. And it explains a structural feature of the German system that survives today: the money is collected and administered by nonprofit sickness funds, not by the state, which is why Germans do not describe their system as government health care.

Sixty-five years later, on 5 July 1948, Britain did something quite different. The National Health Service opened, funded from general taxation, employing the hospital doctors and owning the hospitals. Canada took a third route, and Taiwan a fourth, and each of them achieved universal coverage. This lesson lays the four routes side by side, because the most useful thing comparative health policy teaches is that the goal is reachable by several roads, and that each road has a toll.

Key idea: Universal coverage has been achieved through at least four structurally different financing models, so any claim that a particular institutional arrangement is necessary for it is contradicted by an existing country.

The four models

Model	Who insures	Who provides	Money comes from	Archetype
Beveridge	Government	Mostly government	General taxation	United Kingdom, Spain, Nordic countries
Bismarck	Nonprofit sickness funds, regulated	Mostly private, often nonprofit	Payroll contributions from workers and employers	Germany, France, Japan, Belgium
National health insurance	Government, single payer	Mostly private	Taxation or a dedicated premium	Canada, Taiwan, South Korea
Out of pocket	Nobody, for most people	Whoever is available	Households at the point of service	Much of the low-income world

Two cautions before you use this taxonomy. First, no country is a pure type. Britain has a private sector, Germany has a private insurance tier, Canada leaves most outpatient drugs and dentistry outside its public plan. Second, the United States is not a fifth type; it is all four at once, sorted by demographic group. Veterans Affairs is Beveridge, complete with government-employed physicians and government hospitals. Employer coverage is a mutation of Bismarck. Medicare is national health insurance for people over 65. And the uninsured face an out-of-pocket system. That observation, which T. R. Reid made memorably, is the fastest way to explain American complexity to someone who has not studied it.

Britain: the tax-funded service

The NHS was built on the 1942 Beveridge Report's proposal for a comprehensive health service free at the point of use, and Aneurin Bevan brought the hospital consultants along by, in his own phrase, stuffing their mouths with gold: allowing them to keep private practice alongside NHS salaries. General practitioners were never employed by the NHS at all; they remained independent contractors, paid largely by capitation for a registered list of patients, and they serve as gatekeepers to specialist care.

The system's characteristic strengths are equity and administrative simplicity. There are no bills, no networks, no eligibility determinations, and administrative costs are a fraction of American levels. Its characteristic weakness is what a fixed budget does when demand exceeds it. The English elective waiting list rose above seven million in the 2020s, and waiting is the mechanism by which a globally budgeted system allocates scarce elective capacity.

Britain also does something openly that other countries do quietly. The National Institute for Health and Care Excellence appraises new treatments against cost-effectiveness thresholds expressed in cost per quality-adjusted life year, and it says no in public, with published reasoning. The thresholds have historically sat in the range of twenty to thirty thousand pounds per quality-adjusted life year, with flexibility for end-of-life and highly specialized treatments. That transparency is unusual and politically costly. Every system refuses some treatments; NICE writes down why.

What matters here: A tax-funded national service buys equity and administrative simplicity and pays for it in waiting for elective care and in explicit, public refusals of expensive treatments.

Germany: many funds, one rulebook

Germany's statutory health insurance covers the great majority of the population, with a private tier available to high earners, the self-employed, and civil servants. Membership is with one of around a hundred competing nonprofit sickness funds, consolidated from more than a thousand in 1990. Contributions are a percentage of gross wages, split between employee and employer, with an additional fund-specific rate.

The crucial machinery is invisible to patients. Because funds compete for members and cannot reject anyone, they would otherwise have every reason to chase the healthy. A morbidity-based risk structure compensation scheme redistributes money among funds according to the health of their members, which is the largest working example anywhere of the risk adjustment you met in Lesson 2.

Prices and benefits are set not by the government directly but through what the Germans call joint self-government: a federal committee of sickness fund associations, physician associations, and hospital associations, with patient representatives, that decides what is covered. Patients see specialists without referral, choose freely among physicians, and pay small copayments. The system is more expensive than Britain's, at close to twelve percent of national output, and it buys shorter waits and more choice with the difference.

Canada and Taiwan: two single payers

Canada's system began provincially. Saskatchewan under Tommy Douglas introduced universal hospital insurance in 1947 and universal medical insurance in 1962, the latter provoking a doctors' strike that lasted twenty-three days. Federal cost-sharing legislation spread the model nationally in 1957 and 1966, and the Canada Health Act of 1984 set the five conditions a provincial plan must meet to receive federal money: public administration, comprehensiveness, universality, portability, and accessibility, the last of which effectively bars charging patients for insured services.

What Canada shows is that the model's boundaries matter as much as the model. Physician and hospital services are covered without charge. Outpatient prescription drugs, dentistry, and vision are largely outside the public plan, financed by a patchwork of employer coverage, provincial programs, and household spending, which is why Canadians debate pharmacare and Britons do not. Waiting times for elective procedures and specialist consultations are the standing complaint and were the basis of a 2005 Supreme Court of Canada decision holding that Quebec's ban on private insurance for publicly insured services violated the province's own charter where waits were excessive.

Taiwan built its system from scratch and recently enough to have studied everyone else's. Before 1995 roughly half the population was uninsured under a fragmented occupational scheme. On 1 March 1995 a single-payer National Health Insurance replaced it, and coverage reached almost the entire population within about a year. Enrollees carry a smart card holding their identity and claims history. Administrative costs run at a level American analysts find hard to believe, in the low single digits of total spending, achieved by having one payer, one claims format, and one price schedule. Since 2002 the system has operated under global budgets by sector. Its persistent problem is the mirror image of America's: fees are low, providers see very high patient volumes, and financing the system as the population ages is a recurring political fight.

The upshot: Canada and Taiwan use the same single-payer model to opposite effect on administration and on scope, which shows that what a system covers is a separate decision from how it is financed.

Reading them against each other

The Commonwealth Fund periodically compares wealthy countries on access, equity, care process, administrative efficiency, and health outcomes. Its recent editions place the United States last overall among the countries studied while spending by far the most, with Australia, the Netherlands, and the United Kingdom near the top. The United States scores comparatively well on one dimension, the care process measures covering preventive care and patient engagement, and poorly on the rest.

The single most useful generalization from the comparison is this: every system rations, and they differ in the mechanism rather than in whether they do it.

• Britain and Canada ration chiefly by waiting time for elective care, within a budget.
• Germany and France ration chiefly by price regulation and negotiated fee schedules, accepting higher spending in return for shorter waits.
• Taiwan rations by global budget and low fees, which shows up as very short consultations.
• The United States rations chiefly by price and insurance status: care is fast and abundant for the well insured and unaffordable or unreachable for others.

Notice that this is not a neutral tie. Rationing by queue makes a person wait for a hip replacement; rationing by ability to pay makes a person not have one. Reasonable people weigh those differently, and the weighting is a values judgment rather than an empirical one. What the comparison does settle is that no country has escaped the choice.

Common misconceptions

• Universal coverage requires government-run health care. Germany, the Netherlands, and Switzerland achieve it through regulated nonprofit or private insurers, and Canada achieves it with public financing and private delivery.
• Socialized medicine describes every universal system. Government ownership of hospitals and employment of physicians is the Beveridge model specifically, and even Britain leaves general practice to independent contractors.
• Other countries do not ration. All of them do. The mechanisms differ: queue, budget, negotiated price, explicit refusal on cost-effectiveness grounds.
• Single payer means one government provider. It means one insurer. Canadian physicians are overwhelmingly in private practice, and Taiwanese hospitals include many private institutions.
• The United States is a market system and the others are not. American medical prices are heavily administered through Medicare and propagate outward, while several universal systems rely on competition among insurers or providers.
• Every universal system covers everything. Canada leaves most outpatient drugs and dentistry outside the public plan, which is why pharmacare is a live Canadian argument.

Summing up

• Germany's 1883 Health Insurance Act was the first national scheme, introduced by Bismarck to undercut the socialist movement, and it established the sickness fund structure Germany still uses.
• The four models are Beveridge, Bismarck, national health insurance, and out of pocket; no country is pure, and the United States contains all four sorted by group.
• The NHS, opened in 1948, is tax funded with GPs as independent gatekeepers, and it rations elective care by waiting while refusing treatments explicitly through NICE appraisal.
• German statutory insurance runs through around a hundred competing nonprofit funds, held together by morbidity-based risk compensation and joint self-government.
• Canada's Canada Health Act of 1984 sets five conditions and leaves outpatient drugs and dentistry largely outside the public plan.
• Taiwan's single payer, launched in 1995, reached near-universal coverage within a year with administrative costs in the low single digits.
• Every system rations; they differ in whether the mechanism is a queue, a budget, a negotiated price, or ability to pay.

Sources

1. Commonwealth Fund. (n.d.). International Health Care System Profiles. commonwealthfund.org
2. National Institute for Health and Care Excellence. (n.d.). NICE. nice.org.uk
3. Encyclopaedia Britannica. (n.d.). National Health Service. britannica.com
4. Organisation for Economic Co-operation and Development. (n.d.). Health. oecd.org
5. Reid, T. R. (2009). The Healing of America: A Global Quest for Better, Cheaper, and Fairer Health Care. Penguin Press.

A five dollar fine in Cambridge

On 27 February 1902 the board of health of Cambridge, Massachusetts, adopted a regulation requiring every inhabitant of the city to be vaccinated or revaccinated against smallpox. Henning Jacobson, a Lutheran pastor who had emigrated from Sweden, refused. He said vaccination in childhood had made him ill for years and that one of his sons had been harmed too. Massachusetts had a statute letting local boards order vaccination when they judged it necessary for public health, and Cambridge prosecuted him under it. He was convicted and fined five dollars.

Jacobson took the five dollars to the Supreme Court of the United States, which ruled against him on 20 February 1905 by seven votes to two. Justice John Marshall Harlan wrote that constitutional liberty is not freedom from every restraint, and that a community may defend itself against an epidemic threatening its members.

The loose version of Jacobson stops there. Harlan kept going. The power had to be exercised for health and not arbitrarily; a court could strike down a measure bearing no real relation to protecting health, or one that was a plain invasion of a right; and the opinion contemplated that a person whom vaccination would genuinely endanger should not be forced. Twenty-two years later, in Buck v. Bell, the Court cited Jacobson while upholding the compulsory sterilization of Carrie Buck, on the reasoning that a state able to compel vaccination could compel a surgical operation. Tens of thousands of sterilizations followed under state eugenics statutes. That lineage is the standing argument for keeping public health powers narrow, written down, and reviewable from outside the agency.

Key idea: The founding case of American public health law grants real coercive power and in the same breath names the conditions on it, and the abuse later built on it is exactly why those conditions matter.

The power the states never handed over

The Constitution does not mention health. What it does is leave to the states everything it did not give to Congress, and among the things the states kept is the police power: the general authority to regulate for health, safety, morals, and welfare. Colonial ports were running quarantine before there was a federal government to run one. When Chief Justice Marshall described the limits of the commerce power in Gibbons v. Ogden in 1824, he listed inspection laws and quarantine and health laws as examples of the state authority Congress does not displace.

So the front-line authority for almost everything you would recognize as public health is the state's, delegated by statute to a health department and often delegated again to counties and cities. Closing a restaurant after an inspection. Licensing a nurse, a pharmacy, a tattoo parlor. Ordering isolation. Requiring measles vaccination for school entry. Issuing a boil water order. Registering every birth and death. None of it runs on federal power.

The local layer is easy to underestimate. The National Association of County and City Health Officials counts roughly three thousand local health departments, and their legal relationship to the state differs by state. In centralized states the local office is a unit of the state agency and takes direction from it. In decentralized states it answers to a county commission or a mayor, and the state health officer can advise but not order. That is why national coverage of a local public health fight so often makes no sense: the reporter is describing a chain of command that does not exist there.

How Washington acts when it cannot command

Congress has real health powers, but it hit a wall in Printz v. United States in 1997, where the Court held that the federal government may not conscript state officers to administer a federal program. Washington therefore works through four levers, and knowing which lever is in use tells you most of what you need to know about how far a federal health action can go.

Lever	How it works	Examples	Where it stops
Money with conditions	Requirements attached to grants a state may decline	Immunization and preparedness grants, and above all the Medicare and Medicaid conditions of participation	Dole (1987) allowed conditions; NFIB v. Sebelius (2012) found a limit when refusal costs so much it is not a real choice
Regulation of commerce, by statute	Federal rules on goods and interstate movement	FDA authority over drugs, devices, food and, since 2009, tobacco; section 361 of the Public Health Service Act of 1944	The rule must fit the statute Congress wrote, which is where the eviction moratorium failed in 2021
Direct federal operation	The government runs the program itself	Indian Health Service, Veterans Health Administration, quarantine stations at ports of entry	Reaches only the population inside the program
Information and recommendation	Surveillance, publication, advisory committees	The Morbidity and Mortality Weekly Report, the Advisory Committee on Immunization Practices	Compels nobody directly, though recommendations trigger coverage rules and school requirements

Notice what is missing from that table. The Centers for Disease Control and Prevention, created in Atlanta in 1946 as the Communicable Disease Center out of a wartime malaria program, is a funder, a laboratory network, an epidemiological workforce, and a publisher. It is not a police force. Disease reporting shows what that means in practice. Each state decides by law what clinicians and laboratories must report inside its borders, the national notifiable conditions list is assembled from Council of State and Territorial Epidemiologists position statements, and state reporting to the CDC is voluntary. The national picture of American infectious disease is, legally, an act of cooperation.

The upshot: Federal public health power is mostly the power to pay, to publish, and to regulate products in commerce, so a federal action that looks like a command is usually a condition on money or a recommendation that a state has chosen to adopt.

Six tools, ordered by what they take from you

Public health practice runs on a small toolkit, and the useful way to hold it is as a ladder from least to most coercive. The ethical principle and the litigation doctrine converge here: use the least restrictive means that will actually work, and be able to say why the rung below would not have.

• Information. Calorie labels on chain menus, warning labels, the inspection grade posted in the window. Cheap, popular, usually the weakest in effect.
• Surveillance. Notifiable disease reporting, cancer and birth defect registries, syndromic feeds from emergency departments, wastewater sampling. Legally quiet until it touches identifiable data, where privacy law takes over.
• Taxation and subsidy. Tobacco and alcohol excise, sugary drink taxes in a handful of cities, subsidized produce in nutrition assistance. It moves the price without banning anything, and is unpopular for the same reason.
• Design of products and places. Vehicle safety standards, lead out of paint and pipes, fluoridated water, smoke-free workplaces, building codes. Historically the highest-yield rung by a wide margin, because it changes the default and asks nothing of the individual.
• Licensure and inspection. Restaurant grades, pool permits, clinical licensing, nursing home surveys. Coercive but routine, enforced through a permit rather than by force.
• Direct restraint of persons. Isolation, quarantine, closure orders, compulsory examination. Used rarely, hedged with the most procedure.

Two of those terms get swapped constantly, including by officials who should know better. Isolation separates people known to be infected. Quarantine separates people who were exposed and might be incubating, so it restrains someone who may be perfectly healthy, and its procedure is stricter for that reason: notice of the basis, a hearing before a neutral decision maker, counsel, periodic review. After the anthrax letters of 2001 a team led by Lawrence Gostin produced the Model State Emergency Health Powers Act to standardize those protections, and it was attacked at once from both sides, as a civil liberties threat and as an expansion of authority that already existed.

Where the limits actually bite

Modern courts rarely tell a health agency that its science is wrong. They tell it that somebody else was supposed to make this decision. Three cases make the pattern visible.

In September 2012 the New York City Board of Health capped sugary drink servings in restaurants and stadiums at sixteen ounces. In June 2014 the state's highest court affirmed the rule's annulment on separation-of-powers grounds: the board had made a policy choice belonging to the elected city council. The instructive part is that the same board's 2006 trans fat restriction and its 2008 calorie posting rule both survived, and the calorie rule was later written into federal law by the Affordable Care Act. The board did not lose because sugary drinks are harmless. It lost because it was legislating.

In August 2021 the Supreme Court ended the CDC's residential eviction moratorium, holding that section 361, whose examples run to inspection, fumigation, and destruction of infected animals, could not be stretched to the landlord and tenant relationship nationwide. Not a ruling that eviction is unrelated to transmission. A ruling that Congress had not delegated this.

Then the pair decided on the same day in January 2022. The Court stayed the Occupational Safety and Health Administration's vaccinate-or-test standard for large employers, since a workplace safety statute does not carry a general public health measure aimed at a hazard people meet everywhere. The same day it allowed the Centers for Medicare and Medicaid Services to require vaccination of staff at facilities taking Medicare and Medicaid money, because attaching health and safety conditions to federal payment is what that statute has always done. Two vaccination requirements, opposite results, and the difference was which lever from the table above was being pulled.

What matters here: When a public health measure loses in court, the holding is usually that the agency lacked delegated authority rather than that the measure was unwise, so the remedy is almost always a statute rather than better evidence.

Three cents of the health dollar

Government public health activity is around three percent of national health expenditure in the CMS accounts you took apart in Lesson 10. That is the whole apparatus: surveillance, immunization, restaurant and water inspection, vital records, laboratories, epidemiologists, preparedness.

Set that against what the sector has to show. American life expectancy rose from roughly forty-seven years in 1900 to roughly seventy-seven by 2000, and clinical medicine is not where most of that came from. David Cutler and Grant Miller, writing in Demography in 2005, tracked the arrival of water filtration and chlorination in major American cities and estimated that clean water technologies explained about half of the total mortality decline in those cities between 1900 and 1936, around three quarters of the fall in infant mortality, and nearly two thirds of the fall in child mortality. One engineering intervention, financed by municipal bonds and enforced by health departments, outran a century of therapeutics on those measures.

So why does the budget not reflect it? Part of the answer is Geoffrey Rose's, in a 1985 paper in the International Journal of Epidemiology. Rose distinguished a high-risk strategy, which finds the people most likely to be harmed and treats them, from a population strategy, which shifts the whole distribution a little. Most cases of most common conditions arise not among the few at extreme risk but among the very many at modest risk, so the population strategy prevents more disease and delivers almost nothing to any individual participant. That is the prevention paradox. Nobody sends a thank you letter for the epidemic they did not get, and the result is a sector funded in spikes after emergencies and starved between them. Workforce surveys run by the de Beaumont Foundation and the Association of State and Territorial Health Officials have repeatedly found large shares of staff intending to leave within a few years, and departures accelerated after 2020.

The argument after 2020

Between 2020 and 2023 legislatures in a majority of states rewrote the law on public health emergency powers, a wave tracked by the Network for Public Health Law and its partners. Some statutes cap the length of an executive declaration, some require legislative ratification to extend one, some move authority from health officers to elected boards, some prohibit particular measures outright. Both sides of the argument are holding real evidence.

The case for reasserting legislative control does not depend on denying that the pandemic was serious. It rests on statutes written for a hurricane or a few weeks of an outbreak being used to govern for two years, with orders reaching schools, businesses, worship, and travel, issued by officials who face no election. Several blunt measures rest on weak or contested evidence, including the length of school closures and restrictions on outdoor activity, and some were held after the evidence had turned. Trust in public health institutions fell measurably across the period, and trust is not a soft outcome: it decides whether the next recommendation is followed.

The case for preserving fast executive authority is the arithmetic of exponential growth. When cases double every few days, the time taken to convene a legislature is not a procedural cost but a multiplier. Richard Hatchett and colleagues, in the Proceedings of the National Academy of Sciences in 2007, compared American cities in the 1918 influenza pandemic and found that cities acting earlier, with more overlapping interventions, had substantially lower peak death rates. The other half of this case is about drafting: many new statutes restrict not only emergency orders but routine authority nobody was arguing about, including school immunization enforcement and closure of a contaminated facility.

What would settle it? Partly nothing, because the core disagreement is about who should decide under uncertainty, which is a constitutional question rather than an epidemiological one. Part of it is answerable: which measures bought how much benefit at what cost, and whether narrowing routine authority has degraded outbreak investigation and immunization coverage. Those studies can be done, and the record is still being assembled.

Common misconceptions

• The CDC can order a lockdown. It has essentially no direct authority over people or businesses. It recommends, funds, and publishes; states and localities order.
• Public health law is mostly about epidemics. Almost all of it is routine and permanent: inspection, licensure, water, lead, vital records, registries.
• An emergency declaration creates power. It activates powers a statute already contains, which is why the wording of the statute decides what a governor can do.
• Quarantine and isolation are the same thing. Isolation separates the known infected; quarantine separates the exposed but not known infected, and gets stricter procedure for it.
• Courts strike down health rules because judges reject the science. The recent losses turned on delegated authority, which is why the remedy is a new statute rather than a better study.

What you now know

• Jacobson v. Massachusetts (1905) upheld a five dollar fine for refusing smallpox vaccination and in the same opinion stated the limits: no arbitrary use, a real relation to health, no forcing someone the vaccine would endanger.
• Public health authority is chiefly the states' police power, delegated to state agencies and roughly three thousand local health departments whose relationship to the state varies by state.
• Federal power runs through four levers: conditioned money, regulation of commerce under statutes such as section 361, direct federal programs, and information.
• The toolkit runs from information and surveillance through taxation and product design to licensure and, at the top, restraint of persons, governed by the least restrictive effective means.
• Recent losses for health agencies, from the portion cap to the eviction moratorium to the OSHA standard, turned on who was authorized to decide.
• Public health is about three percent of health spending, while Cutler and Miller credited clean water alone with roughly half the urban mortality decline from 1900 to 1936.
• Rose's prevention paradox explains the political weakness of population measures, and most states rewrote emergency powers law after 2020 in an argument setting speed under exponential growth against democratic control.

Sources

1. Supreme Court of the United States. (1905). Jacobson v. Massachusetts, 197 U.S. 11. supreme.justia.com
2. Rose, G. (1985). Sick individuals and sick populations. International Journal of Epidemiology, 14(1), 32-38. doi.org
3. Centers for Disease Control and Prevention. (n.d.). Centers for Disease Control and Prevention. cdc.gov
4. Network for Public Health Law. (n.d.). Network for Public Health Law. networkforphl.org
5. Gostin, L. O., and Wiley, L. F. (2016). Public Health Law: Power, Duty, Restraint (3rd ed.). University of California Press.

Eight volumes, and sixty thousand deaths a year

In 1984 Margaret Heckler, secretary of Health and Human Services under Ronald Reagan, was reading a departmental health report and noticed something her staff could not explain: Black Americans were dying at higher rates than white Americans, and nobody in the department could say by how much, from what, or why. She commissioned a task force. The Report of the Secretary's Task Force on Black and Minority Health, published in 1985 in eight volumes, put a number on it. Compared with the death rates of white Americans, minority populations were experiencing roughly sixty thousand excess deaths a year, concentrated in six causes: cardiovascular disease and stroke, cancer, infant mortality, diabetes, cirrhosis, and homicide and unintentional injury.

The number was not the point. The point was that a cabinet department discovered, in the middle of the 1980s, that it had never measured this. The Heckler Report created the federal Office of Minority Health, made the excess death count a standing statistic, and set the shape of everything that followed: measure the gap, name the causes, argue about the remedy.

Why this matters: The American record on unequal health is short because the measurement is recent, and much of what sounds like a permanent finding is a series that started inside living memory.

A difference is not automatically an injustice

Men die younger than women in essentially every country. That is a difference in health between groups, and almost nobody calls it an injustice requiring policy remedy. Eighty-year-olds use more hospital care than twenty-year-olds. Also a difference, also not a grievance.

Margaret Whitehead drew the line that the field now uses, in a 1992 paper in the International Journal of Health Services. A disparity is any measured difference between groups. An inequity is the subset of differences that are avoidable and unfair: they arise from arrangements that could be otherwise, and that we would not defend if asked to defend them out loud. A gap in life expectancy caused by unequal exposure to lead in drinking water is an inequity. A gap caused by biological sex is not, because it is not avoidable in any meaningful sense.

Keep the two words separate when you read. A study reporting a disparity has done a measurement. A claim of inequity has added a judgment about avoidability and fairness, and that judgment can be argued with on its own terms rather than by disputing the measurement.

Access is five things, and coverage is only one

Roy Penchansky and William Thomas, writing in Medical Care in 1981, broke access into five components that still hold up. Their point was that a system can fix one and leave a person exactly as unable to get care as before.

Component	The question it asks	What failure looks like
Availability	Does the service exist in sufficient supply?	No obstetric unit within eighty miles
Accessibility	Can the person physically get there?	Two bus transfers and no evening service
Accommodation	Is it organized to fit the person's life?	Appointments only between nine and four, no walk-ins, phone tree only
Affordability	Can the person pay what it costs them?	Insured, but a 4,000 dollar deductible on an hourly wage
Acceptability	Will the person use it, and will the clinic take them?	No interpreter, or a practice that has closed its panel to Medicaid

Read the affordability row again. By the mid-2020s the uninsured share of the American population had fallen to roughly eight percent, the lowest ever recorded, with something like twenty-six million people uninsured. That is a genuine achievement of the coverage expansions in Lesson 5. It is also compatible with a very large number of insured people who cannot afford care, because deductibles in employer plans have grown far faster than wages. Underinsurance is an affordability failure inside a coverage success, and any account that measures access by the insured share will miss it entirely.

The point: Coverage is one of five conditions for access, so a policy that raises the insured share can leave availability, accommodation, and real affordability untouched.

The gaps, with dates attached

Numbers here move, and a figure without a year is not a fact. These are the ones worth carrying, each with its source and its vintage.

• Infant mortality. The national rate was about 5.6 deaths per 1,000 live births in 2022 in the National Center for Health Statistics data. For infants of non-Hispanic Black mothers it was roughly 10.9; for infants of non-Hispanic white mothers, roughly 4.5. The ratio has been close to two to one for as long as the series has existed.
• Income and life expectancy. Raj Chetty and colleagues, using tax and mortality records for 2001 to 2014, found a gap in life expectancy at age forty of 14.6 years between the richest and poorest one percent of American men, and 10.1 years for women.
• Place. The same study found that life expectancy among the poorest Americans varied by four to five years across metropolitan areas, and that the variation correlated with local behaviors and local conditions rather than with measures of access to medical care.
• The coverage gap. In the states that have not adopted the Medicaid expansion, adults too poor for marketplace subsidies and ineligible for Medicaid number well over a million on KFF's periodic estimates, a count that falls each time a state adopts.

Two mechanisms that are not about attitudes

Ask most people how unequal care happens and they will describe a prejudiced clinician. That happens, and the Institute of Medicine's 2003 report Unequal Treatment assembled a large body of evidence that differences in care persisted after accounting for insurance, income, and clinical need. But two of the best documented mechanisms run through equipment and arithmetic, which makes them both less morally satisfying and much easier to fix.

The first is the pulse oximeter, the clip on the finger that reads blood oxygen from how light passes through tissue. Michael Sjoding and colleagues, writing in the New England Journal of Medicine in December 2020, compared oximeter readings against arterial blood gas measurements in hospitalized patients. Occult hypoxemia, meaning true arterial saturation below 88 percent while the oximeter read a reassuring 92 to 96 percent, occurred in Black patients about three times as often as in white patients. The device was reading skin pigment as if it were oxygen. During a respiratory pandemic, a systematically optimistic reading is a systematically delayed escalation of care.

The second is an equation. For decades laboratories estimating kidney function from serum creatinine applied a multiplier for patients identified as Black, raising their reported estimated glomerular filtration rate. A higher reported number means less severe kidney disease, which means later referral to a nephrologist, later dialysis planning, and later placement on a transplant list. In 2021 a joint task force of the National Kidney Foundation and the American Society of Nephrology recommended immediate adoption of a race-free equation, and laboratories converted.

Worth holding on to: Neither the oximeter nor the kidney equation required anyone to hold a prejudiced belief, which is why fixing unequal care is partly an engineering and standards problem rather than only a training problem.

The second address problem

Where you live decides which of Penchansky's five components fail for you. The Cecil G. Sheps Center at the University of North Carolina counted more than one hundred rural hospital closures between 2010 and 2020, and closures fall hardest on obstetrics and inpatient psychiatry, the service lines with the thinnest margins. March of Dimes has classified more than a third of American counties as maternity care deserts, meaning no hospital offering obstetric care and no obstetric provider. A pregnant woman in such a county is not underinsured. She is un-served, and no insurance card creates a delivery unit.

The federal instrument here is the shortage designation. The Health Resources and Services Administration designates health professional shortage areas and medically underserved areas, and those designations pull in the National Health Service Corps, which repays student loans for clinicians who practice in them, and support federally qualified health centers, which serve patients regardless of ability to pay on a sliding scale. Those programs are small relative to the problem and are among the best evidenced things in the sector, which is an uncomfortable combination.

Where the argument actually is

Nobody serious disputes that the gaps exist. The dispute is about where a marginal dollar does the most good, and both camps hold real evidence.

The case for coverage first rests on quasi-experimental mortality evidence that has strengthened over time. Sarah Miller, Norman Johnson, and Laura Wherry, publishing in the Quarterly Journal of Economics in 2021, linked survey responses to administrative death records and compared low-income adults in states that expanded Medicaid with those in states that did not. They found a reduction in annual mortality of roughly nine percent among those most likely to gain coverage, concentrated in deaths from causes medicine can address. Benjamin Sommers and colleagues had found a comparable direction in the New England Journal of Medicine in 2012 for earlier state expansions. Against this, the Oregon lottery from Lesson 2 found no statistically significant effect on measured blood pressure, cholesterol, or blood sugar at two years, though it did find large reductions in depression and in financial catastrophe. The honest summary is that coverage clearly protects people financially and reduces depression, and the mortality evidence is real but comes from designs weaker than a randomized trial.

The case for upstream investment starts from an accounting observation and one experiment. Elizabeth Bradley and colleagues, writing in BMJ Quality and Safety in 2011, noted that the United States is not an outlier in total spending on health plus social services combined; it is an outlier in the ratio, spending far more on medical care relative to housing, income support, and nutrition than peer countries do, and countries with higher social to health ratios showed better outcomes. Then Moving to Opportunity, a genuine randomized experiment in which the Department of Housing and Urban Development gave vouchers to move out of high-poverty neighborhoods in the 1990s. Jens Ludwig and colleagues reported in the New England Journal of Medicine in 2011 that the moves produced measurable reductions in extreme obesity and diabetes more than a decade later, without any medical intervention at all.

What tempers the upstream case is the record of bolting social interventions onto medical delivery. Amy Finkelstein and colleagues randomized the Camden Coalition's intensive care management for very high-utilizing patients, a program widely admired and widely copied, and reported in the New England Journal of Medicine in 2020 that it produced no significant reduction in readmissions. The comparison group improved just as much, because extreme use regresses toward the mean, exactly as Lesson 10 warned.

What would separate the positions is not another cross-country correlation. It is randomized or well-identified tests of specific interventions, with mortality and function as endpoints rather than utilization, and long enough follow-up to catch effects that a two-year window cannot. Those studies are being run now, and until they report, a policymaker choosing between a coverage dollar and a housing dollar is choosing between two evidence bases of different shapes rather than between evidence and ideology.

Common misconceptions

• Insured means able to get care. Coverage addresses one of five access components, and a high deductible on an hourly wage is an affordability failure inside a coverage success.
• Disparity and inequity mean the same thing. A disparity is a measurement; an inequity adds a judgment that the difference is avoidable and unfair.
• Unequal care is mostly about individual prejudice. Two of the best documented mechanisms are a device calibration and a laboratory equation, neither of which required anyone to believe anything.
• Racial gaps in health are explained by income. Gaps in infant mortality and maternal outcomes persist across income and education levels, which is why the research has moved toward exposure, treatment, and place.
• Rural access is a money problem. When the obstetric unit has closed, the constraint is availability, and no payment rate creates a delivery room overnight.
• The evidence settles the coverage versus upstream question. The two literatures use different designs and endpoints, and the best studies on each side are measuring different things.

What to remember

• The 1985 Heckler Report counted roughly sixty thousand excess deaths a year in minority populations and created the federal machinery for measuring them.
• Whitehead's distinction holds: a disparity is a measured difference, an inequity is the avoidable and unfair subset of them.
• Penchansky and Thomas split access into availability, accessibility, accommodation, affordability, and acceptability, and coverage touches mainly the fourth.
• Infant mortality for infants of Black mothers ran roughly 10.9 per 1,000 against roughly 4.5 for white mothers in 2022, a ratio close to two to one throughout the series.
• Chetty and colleagues measured a 14.6 year life expectancy gap between the richest and poorest one percent of men, with local variation among the poor that did not track medical access.
• Pulse oximetry overestimated oxygen saturation in Black patients about three times as often as in white patients, and the race coefficient in kidney function estimation was removed in 2021.
• More than one hundred rural hospitals closed between 2010 and 2020, and over a third of counties are maternity care deserts, which is an availability failure rather than a coverage failure.
• Medicaid expansion reduced mortality by roughly nine percent among likely gainers in the strongest quasi-experimental study; Moving to Opportunity produced health gains from housing alone; Camden's care management produced none.

Sources

1. Chetty, R., Stepner, M., Abraham, S., et al. (2016). The association between income and life expectancy in the United States, 2001-2014. JAMA, 315(16), 1750-1766. doi.org
2. Sjoding, M. W., Dickson, R. P., Iwashyna, T. J., Gay, S. E., and Valley, T. S. (2020). Racial bias in pulse oximetry measurement. New England Journal of Medicine, 383(25), 2477-2478. doi.org
3. Miller, S., Johnson, N., and Wherry, L. R. (2021). Medicaid and mortality: New evidence from linked survey and administrative data. Quarterly Journal of Economics, 136(3), 1783-1829. doi.org
4. KFF. (n.d.). Uninsured. kff.org
5. Institute of Medicine. (2003). Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care. National Academies Press.

Sixty votes, and then fifty-nine

On the morning of 24 December 2009 the Senate passed its version of health reform sixty votes to thirty-nine. Sixty was the exact number needed to end debate, so every one of them was load bearing. The plan from there was ordinary: a conference committee with the House, a merged bill, another vote in each chamber.

On 19 January 2010 Massachusetts held a special election to fill the seat left by Edward Kennedy's death, and Scott Brown won it. The sixtieth vote was gone, and with it the ability to pass anything new through the Senate. What survived was the bill the Senate had already passed. So on 21 March 2010 the House swallowed the Senate's text whole, 219 votes to 212, without changing a comma, and the president signed it on 23 March. The fixes the House wanted were packaged separately into a reconciliation bill, which needs only fifty-one votes, and signed on 30 March.

That sequence explains features of the Affordable Care Act that otherwise look like drafting errors. Odd cross-references, provisions that read like a first draft, a phrase about an exchange established by the state that later carried a Supreme Court case: those are artifacts of a bill that was never conferenced. Legislative procedure is not the packaging around policy. It leaves fingerprints on the text you eventually have to implement.

In short: A law is shaped as much by the arithmetic of the chamber that passed it as by what its authors wanted, and reading a statute without knowing its procedural history will mislead you.

Where a bill comes from before anyone drafts it

John Kingdon's study of federal agenda setting, first published in 1984, remains the most useful account of why some problems get solved and others sit for decades. Kingdon described three streams flowing independently. The problem stream is what people have come to see as a problem, often because an indicator moved or a focusing event happened. The policy stream is the supply of worked-out proposals circulating among specialists, most of them years old. The politics stream is elections, party control, and the national mood.

Nothing happens while the streams run separately. A window opens when they join, usually for a short time, and a policy entrepreneur who has kept a proposal ready couples them. The combination at the center of the ACA, guaranteed issue plus a coverage requirement plus subsidized private plans in a regulated market, was not invented in 2009. It had circulated among health economists since the late 1980s, appeared in a Republican alternative to the Clinton plan in 1993, and was enacted in Massachusetts in 2006. It was sitting on the shelf when the window opened.

Committees, and the scorekeeper everyone fears

A health bill does not go to a health committee, because there is no such thing. Jurisdiction is split by what the bill touches. In the House, Ways and Means holds Medicare Part A and the tax code, Energy and Commerce holds Medicaid and Part B and the Public Health Service Act, and a third committee holds employer benefit law. In the Senate, Finance holds the money and HELP holds the rest. A comprehensive bill therefore gets written in pieces by committees that do not agree, which is why the seams show.

Then the Congressional Budget Office. CBO estimates what a bill does to federal spending and revenue over a ten-year window, relative to a baseline that assumes current law continues. The estimate is not a prediction anyone believes precisely. It is a scorekeeping convention, and because budget rules attach consequences to the number, the number changes the bill. In March 2010 CBO and the Joint Committee on Taxation estimated the ACA would reduce deficits by about 143 billion dollars over its first decade, and that figure was fought over line by line for a year.

Two consequences follow from the ten-year window, and both are visible in real statutes. Provisions get delayed so that costs fall outside the window. Revenue gets pulled forward. And a program that is authorized is not thereby funded: authorization creates the legal power to run a program, appropriation supplies the money, and the graveyard of health policy is full of authorized programs that never received a dollar.

The core of it: The scorekeeper's rules are not neutral plumbing, because a ten-year budget window rewards delaying costs and accelerating revenue, and statutes get drafted to exploit exactly that.

Reconciliation, and the rule named after a senator

Budget reconciliation was created by the Congressional Budget Act of 1974 to let Congress bring spending and revenue into line with a budget resolution. Debate is limited, so a filibuster is impossible, and a bill passes the Senate with a simple majority. That makes it the only route for major legislation when a party holds the Senate but not sixty votes, which is most of the time.

The constraint is the Byrd rule, added in the 1980s and named for Robert Byrd, which strips provisions deemed extraneous to the budget. A provision fails if it produces no change in outlays or revenues, or if the budgetary effect is merely incidental to a policy change. The Senate parliamentarian advises on each challenge, and those rulings decide what a bill can contain. This is why so much American health policy arrives as a tax provision or a payment rate change rather than as a regulatory command: the tax version survives reconciliation and the regulatory version does not. The drug price negotiation provisions of 2022 were structured the way they were partly for this reason.

The statute is not the rule

A signed bill is an instruction to an agency. Turning it into something a hospital can comply with runs through the Administrative Procedure Act of 1946, and the sequence is fixed.

1. The agency drafts a proposed rule and publishes it in the Federal Register with its reasoning and its estimates.
2. A comment period opens, typically thirty to sixty days. Anyone may comment, and major health rules routinely draw thousands of comments from hospitals, insurers, manufacturers, physician societies, and patient groups.
3. The agency issues a final rule that must respond to significant comments. Failing to respond is one of the main ways a rule loses in court.
4. Significant rules pass through the Office of Information and Regulatory Affairs inside the Office of Management and Budget, a review established in its current form by an executive order in 1993, where costs and benefits are examined and other agencies weigh in.
5. The rule takes effect after a delay, and Congress has a window during which a joint resolution under the Congressional Review Act of 1996 can void it by simple majority in both chambers.

Agencies also issue guidance, manuals, frequently asked questions, and letters to state Medicaid directors, none of which go through notice and comment. Guidance is faster and legally weaker, and the boundary between an interpretive document and a binding rule is one of the most litigated questions in administrative law.

Then the courts, where the ground moved recently. For forty years, under the Chevron doctrine of 1984, a court that found a statute ambiguous generally deferred to the agency's reasonable interpretation. In 2024 the Supreme Court overruled Chevron in Loper Bright Enterprises v. Raimondo, holding that courts must determine the best reading of a statute themselves. Alongside it sits the major questions doctrine, named in a 2022 case, under which an agency claiming power of vast economic and political significance must point to clear congressional authorization. Together they shift leverage from agencies to courts and, ultimately, back to Congress, which now has to write with more precision or watch its rules fall.

The same law, two provisions, two fates

Take two things Congress enacted in the same statute in March 2010 and follow each to its end.

Section 4205 required chain restaurants to post calorie counts. The FDA proposed a rule, took comments, and issued a final rule in December 2014. Industry objected that pizza chains and grocery prepared foods could not comply as written, compliance dates slipped repeatedly, and the requirement finally took effect in May 2018. Eight years from statute to enforcement, and it survived because the agency could actually write a workable rule and a real constituency wanted it.

Title VIII created the CLASS Act, a voluntary public insurance program for long-term care services, financed by premiums from working adults. Long-term care is the enormous gap in American coverage you met in Lesson 4, so the problem was real. But the statute required the secretary to certify that the program would be solvent for seventy-five years, and voluntary insurance against a risk that mainly worries people who expect to need it is the adverse selection problem from Lesson 2 in its purest form. In October 2011 the department announced it could not design a program meeting the statutory test. Congress repealed the title in January 2013. It was law for almost three years and never covered anyone.

The difference was not politics or funding. It was that one provision could be operationalized and the other could not, and nobody discovered that until after enactment. Jeffrey Pressman and Aaron Wildavsky made the general version of this point in 1973 after watching a federal economic development program in Oakland produce almost nothing despite money, consensus, and goodwill: each step in an implementation chain has a probability of clearance below one, and multiplying enough of them together drives the product toward zero.

Remember: Enactment is a milestone, not a finish line, and a provision that cannot be turned into an administrable rule dies quietly years after the signing ceremony.

The website that made the argument concrete

The marketplaces opened on 1 October 2013. The federal website failed almost immediately under load and under its own architecture, and internal notes later released to congressional investigators recorded six enrollments completed on the first day. A rescue team rebuilt the site over roughly two months; the first open enrollment period ended in spring 2014 with about eight million marketplace sign-ups.

The failure was not a policy failure and not a legal one. It was procurement, systems integration, and the absence of anyone with authority over the whole thing. That is Michael Lipsky's point from a different angle: in his 1980 account of street-level bureaucracy, the people at the counter, in the call center, and at the eligibility desk make the policy real, and their working conditions and discretion determine what citizens actually experience. A statute, a rule, and a functioning queue are three separate accomplishments.

Common misconceptions

• Passing the bill is the hard part. Rulemaking usually takes longer than legislating, and provisions that cannot be operationalized fail after enactment.
• A CBO score predicts what will happen. It is a scorekeeping estimate against a defined baseline over ten years, and its rules shape how bills are drafted.
• Reconciliation lets a majority pass anything. The Byrd rule strips provisions whose budgetary effect is merely incidental, which is why so much health policy arrives as tax or payment changes.
• Agency guidance has the force of a regulation. Guidance skips notice and comment, is legally weaker, and is frequently challenged as a rule in disguise.
• Authorizing a program funds it. Authorization grants the power to act; appropriation supplies the money, and many authorized health programs never received any.
• Courts defer to agency expertise on ambiguous statutes. That was Chevron, overruled in 2024, and the major questions doctrine narrows agency reach further.

Recap

• The Senate passed health reform 60 to 39 on 24 December 2009; after the Massachusetts special election of 19 January 2010 the House adopted the Senate text unchanged, 219 to 212, with fixes carried separately by reconciliation.
• Kingdon's three streams explain timing: proposals sit ready for years, and a window opens when problem, policy, and politics briefly align.
• Health bills are split across committees by jurisdiction, and the CBO score, built on a ten-year window and a current-law baseline, changes what gets drafted.
• Reconciliation avoids the filibuster but the Byrd rule strips provisions whose budgetary effect is incidental, pushing policy into tax and payment form.
• Rulemaking runs proposal, comment, final rule with responses, regulatory review, and a Congressional Review Act window, with guidance as the faster and weaker alternative.
• Loper Bright overruled Chevron in 2024, and courts now determine the best reading of a statute themselves rather than deferring to agencies.
• Menu labeling took eight years from statute to enforcement and worked; the CLASS Act was law for three years and never covered anyone because it could not be made solvent.
• The healthcare.gov launch of 1 October 2013 recorded six completed enrollments on day one and about eight million sign-ups by the end of that first open enrollment.

Sources

1. United States Congress. (n.d.). Congress.gov: Legislation and the legislative process. congress.gov
2. Office of the Federal Register. (n.d.). Federal Register. federalregister.gov
3. Congressional Budget Office. (n.d.). Congressional Budget Office. cbo.gov
4. Kingdon, J. W. (2011). Agendas, Alternatives, and Public Policies (updated 2nd ed.). Longman.
5. Pressman, J. L., and Wildavsky, A. (1984). Implementation (3rd ed.). University of California Press.

The eighty-five percent line

In 1999 Adrian Bagust, Michael Place, and John Posnett published a simulation in the BMJ that every hospital executive should have taped inside a drawer. They modelled a hospital taking emergency admissions with a fixed number of beds and asked a simple question: as average occupancy rises, how does the risk of having no bed for the next arrival behave? The answer was that it does not rise smoothly. Below roughly 85 percent average occupancy the hospital absorbs its bad days. Above it, the curve turns up sharply, and by 90 percent regular crisis is not a management failure but an arithmetic certainty.

This surprises people because the intuition is linear. If you run at 95 percent you have five percent slack, which sounds like enough. It is not, because arrivals are random and lengths of stay are variable, so what matters is not the average but the tail. A queue in front of a nearly full system does not grow a little. It explodes.

That single fact reframes a great deal of what looks like chaos in a hospital: emergency departments boarding admitted patients for hours, elective surgeries cancelled the morning of, ambulances diverted. The system is not badly run. It is run too close to the line, usually because empty beds look like waste on a report.

So what?: Capacity failures in a hospital are usually a queueing problem rather than a staffing scandal, and the fix is to move the operating point rather than to work the existing staff harder.

The variability nobody schedules, and the variability someone does

Ask why the hospital is full on Tuesday and the answer will be the emergency department. Eugene Litvak and colleagues, writing in the Joint Commission Journal on Quality and Patient Safety in 2005, argued the opposite and had the data for it. Emergency arrivals are random, but random in a well behaved way: over a month they follow a predictable distribution, and a manager can plan for a distribution. What is not predictable is the elective surgical schedule, because it is built around surgeons' block times and personal preferences, which pile cases onto Monday and Tuesday and leave Friday light.

Litvak's distinction is between natural variability, which is inherent in illness, and artificial variability, which the organization creates itself and then treats as a fact of nature. Artificial variability is the one that can be removed. Smoothing the elective schedule across the week, and separating scheduled from unscheduled surgical flow so an emergency case does not bump an elective one, lowers the peak census without adding a single bed. Cincinnati Children's Hospital Medical Center applied the method in the mid-2000s and reported moving substantially more surgical volume through the same physical plant.

Notice what makes this hard, and it is not analysis. Smoothing the schedule means telling surgeons, the people who generate the revenue and who are frequently not employees, that their block time is moving. That is the actual job.

What a health care manager is managing

The Bureau of Labor Statistics counts medical and health services managers as one of the fastest growing occupations it tracks, with projected growth in the high twenties of percent over its recent ten-year windows and a median wage a little above 110,000 dollars in its May 2023 estimates. What the title covers is five things at once: money, people, capacity, quality, and a professional workforce that does not sit in the reporting line.

That last item is the structural peculiarity. In a factory, authority runs down an organization chart. In a hospital, the physicians who admit patients, order tests, and decide length of stay are frequently not employed by the hospital at all. They hold privileges granted by the medical staff, a self-governing body with its own bylaws, its own elected officers, and its own credentialing committees, answerable to the board rather than to the chief executive. The chief executive cannot instruct a physician to discharge earlier. They can change what is convenient, what is measured, what is paid for, and what gets discussed at the department meeting. Management by influence is not a soft skill here. It is the operating model.

Reading the money

Open a hospital income statement and the first number is the largest and the least meaningful. Gross charges come from the chargemaster, a list price schedule that almost nobody pays. Subtract contractual allowances, meaning the difference between charges and what each payer has agreed to pay, and you reach net patient revenue, which is the real number. A hospital reporting three billion in charges may be collecting one billion.

Below that sits payer mix, and payer mix is destiny. The same appendectomy generates entirely different revenue depending on the card in the patient's wallet, which means two hospitals with identical clinical operations can have opposite financial results because one sits in a commercially insured suburb and the other serves Medicaid and the uninsured.

Then the operating margin, which is thinner than almost anyone outside the sector expects. The American Hospital Association reported that roughly half of United States hospitals finished 2022 with negative operating margins, the worst year in the modern series, before a partial recovery into the low single digits. MedPAC has reported aggregate hospital margins on Medicare patients running somewhere between negative eight and negative thirteen percent through the early 2020s.

What that Medicare margin means is genuinely contested, and both readings are serious. One reading is straightforward underpayment: Medicare rates are set administratively and have not kept pace with input costs, so hospitals lose money on the largest single payer and make it back on commercial patients. The other reading is MedPAC's own. It notes that hospitals under real financial pressure, without a strong commercial base to lean on, run measurably lower costs per case with comparable quality, which suggests that costs partly expand to absorb available revenue rather than being fixed by the medicine. Austin Frakt, reviewing the evidence in the Milbank Quarterly in 2011, found that the classic cost shifting story, where a dollar lost on Medicare is recovered dollar for dollar from private payers, is much weaker in the data than the sector's rhetoric implies. Both readings agree on the arithmetic and disagree about causality, and a manager who cannot articulate the second one is not ready to negotiate with anybody.

Why cost per case is usually an allocation

Take a joint replacement line. Suppose the direct variable cost of a case, meaning the implant, drugs, supplies, and the incremental staffing time, is 11,000 dollars. The accounting system allocates a further 6,000 dollars of building, equipment, and administration, producing a fully allocated cost of 17,000. Now the payers: a commercial plan pays 28,000, Medicare pays 14,000, Medicaid pays 9,000.

Payer	Payment	Result on fully allocated cost of 17,000	Contribution margin over variable cost of 11,000
Commercial	28,000	Profit of 11,000	Plus 17,000
Medicare	14,000	Loss of 3,000	Plus 3,000
Medicaid	9,000	Loss of 8,000	Minus 2,000

Read the two right-hand columns against each other, because they give opposite instructions. On fully allocated cost the Medicare case loses money and a manager might want less of it. On contribution margin the Medicare case pays 3,000 dollars toward fixed costs that exist whether or not the case happens. Stop doing Medicare joints and the building does not get cheaper; you have simply removed 3,000 dollars per case of contribution and kept the whole 6,000 allocation, now spread across fewer cases.

Change one input and the answer flips. If dropping the volume would let the hospital close an operating room, retire the lease, and redeploy the team, then part of that 6,000 stops being fixed, and the fully allocated view becomes the right one. This is the whole reason Robert Kaplan and Michael Porter argued in the Harvard Business Review in 2011 that health care does not know what anything costs: the allocations that dominate hospital reporting are accounting conventions, not measurements, and decisions made on them are made on fiction.

Bottom line: Ask whether a cost disappears if the activity stops, and if it does not, keep the activity while its contribution margin is positive.

Labor is the budget

Roughly half of a hospital's operating expense is labor, which means every serious cost conversation is a staffing conversation. It also means the two levers a manager has are the number of people and the mix of people, and both are constrained by clinical requirements, licensure, and, in many places, a union contract.

Turnover is the underrated line. Retention surveys in the sector put the cost of replacing one staff nurse in the tens of thousands of dollars once recruitment, orientation, and lost productivity are counted, and the same organizations frequently decline to spend a fraction of that on the conditions that would have kept the nurse. During 2021 and 2022 hospitals spent several times their historic outlay on contract and travel staff at premium rates, which is what a retention failure looks like once it has become an emergency.

Who is actually in charge

Three authorities sit above the operating hospital, and they do different jobs. The board of trustees holds fiduciary duty, hires and fires the chief executive, and, less well understood, carries legal responsibility for quality of care rather than only for finances. Accreditation bodies such as the Joint Commission survey against their own standards, and because the Centers for Medicare and Medicaid Services grants approved accreditors deeming authority, an accredited hospital is treated as meeting the federal conditions of participation without a separate state survey. That is the real leverage: accreditation is voluntary in name and mandatory in effect, because losing it means losing Medicare. And the medical staff governs its own membership through bylaws, credentialing, and peer review.

A manager who wants to change clinical practice therefore has four routes and no direct one: convince the medical staff leadership, change what the payment system rewards, change the default in the ordering system, or bring the board's quality committee into it. Which of those you reach for is most of the craft.

Why the pilot works and the rollout does not

In 2002 Virginia Mason Medical Center in Seattle sent its executives to Japan to study the Toyota Production System and came back to build the Virginia Mason Production System around it: standard work, visual management, rapid improvement events, stopping the line when something goes wrong. It is the most studied Lean implementation in American health care and it produced real results.

The honest summary of the wider literature is less flattering. Systematic reviews of Lean and related methods in health care repeatedly find strong results in individual projects and weak evidence that whole-organization programs sustain measurable gains. That gap is not evidence of fraud. It is the predictable consequence of three things. A pilot runs where a champion volunteered, on a unit that was interested. It gets attention, which changes behavior on its own. And units are usually chosen because they were performing badly, so some of the improvement is regression to the mean, exactly as in Lesson 10.

The practical rule for a manager reading a proposal is to ask three questions of any reported success. Compared with what: was there a control unit or only a before and after? Who chose the site, and were they already keen? And what happened eighteen months later, after the champion moved on and the attention went elsewhere? A method that cannot answer the third question is a project, not a system.

Common misconceptions

• A full hospital is an efficient hospital. Above roughly 85 percent occupancy the queueing behavior turns sharply, so high occupancy buys visible crisis for invisible savings.
• Emergency demand is what makes capacity unpredictable. Emergency arrivals follow a stable distribution; the elective schedule is the variability the organization creates itself.
• Hospital charges tell you what care costs. Charges come from a list price almost nobody pays, and net patient revenue after contractual allowances is a fraction of it.
• A service losing money on fully allocated cost should be cut. If its contribution margin is positive and the fixed costs stay, cutting it makes the hospital worse off.
• The chief executive can direct physicians. Most physicians hold privileges rather than employment, and the medical staff is a self-governing body answering to the board.
• Accreditation is optional. Approved accreditors hold deeming authority for the Medicare conditions of participation, so losing accreditation means losing the largest payer.

Looking back

• Bagust, Place, and Posnett showed in 1999 that the risk of having no bed rises sharply above roughly 85 percent average occupancy, because what matters is the tail rather than the average.
• Litvak's distinction separates natural variability, inherent in illness, from artificial variability created by the elective schedule, and only the second can be removed.
• A manager runs money, people, capacity, quality, and a physician workforce that mostly holds privileges rather than jobs, which makes influence the operating model.
• Gross charges are a list price; net patient revenue after contractual allowances is the real figure, and payer mix largely determines a hospital's financial result.
• About half of American hospitals had negative operating margins in 2022, and MedPAC has reported Medicare margins around negative eight to negative thirteen percent, a figure read as underpayment by one side and as costs expanding to fit revenue by the other.
• Fully allocated cost and contribution margin give opposite instructions, and the deciding question is whether a cost actually disappears when the activity stops.
• Labor is roughly half of operating expense, turnover costs tens of thousands of dollars per nurse, and the contract labor spike of 2021 and 2022 was a retention failure priced at the spot rate.
• Improvement pilots outperform their own rollouts because of champion selection, attention effects, and regression to the mean, so ask what happened eighteen months later.

Sources

1. American Hospital Association. (n.d.). Fast Facts on U.S. Hospitals. aha.org
2. Medicare Payment Advisory Commission. (n.d.). MedPAC reports to Congress. medpac.gov
3. Bureau of Labor Statistics. (n.d.). Medical and health services managers. Occupational Outlook Handbook. bls.gov
4. Bagust, A., Place, M., and Posnett, J. W. (1999). Dynamics of bed use in accommodating emergency admissions: Stochastic simulation model. BMJ, 319(7203), 155-158.
5. Kaplan, R. S., and Porter, M. E. (2011). How to solve the cost crisis in health care. Harvard Business Review, 89(9), 46-52.