Chapter 12: GLP-1 Agonists and Society — Access, Cost, Shortages, Off-Label Use, and the Ethics of a Hundred-Billion-Dollar Drug Class

"The law, in its majestic equality, forbids the rich as well as the poor to sleep under bridges, to beg in the streets, and to steal loaves of bread." — Anatole France, Le Lys Rouge (1894)

Overview

Everything up to here has been solvable with evidence. Chapter 8 told you how semaglutide is built and why the modifications work. Chapters 9 through 11 told you what it does to appetite, to gastric emptying, and to a person's weight over sixty-eight weeks. Those are questions with answers, and the answers are good ones.

This chapter is different, and it is fair to say so at the outset. What a drug should cost, who should pay, who gets it first when there is not enough, whether an obese person is sick, and what a society owes someone whose condition is partly biological, partly behavioral, and partly the predictable output of the environment they were born into — none of these are settled by trials. They are settled, when they are settled at all, by argument, budget, and politics.

That does not make them optional. A drug nobody can obtain has an efficacy of zero in practice, whatever the trial said. A person who starts therapy, benefits, loses coverage at their employer's next plan renewal, and regains the weight has received an expensive demonstration of physiology and no lasting clinical benefit. The evidence literature and the access literature describe the same molecule and disagree about what it is worth, because they measure different things.

So this chapter asks two disciplines of you at once. The first you have had since Chapter 5: separate the claim from the molecule, state what the evidence supports and what it does not, never upgrade a rating with a mechanism, never downgrade one with distaste. The second is newer. When a question is genuinely contested, present each position at its strongest and do not pretend the disagreement away. You will feel that most sharply in §12.9, where this book declines to tell you whether obesity is a disease and instead tells you what each of four serious positions claims and where each is weakest.

One promise. This chapter is about money, bodies, and shame, and those three together produce more bad writing than almost any other combination in medicine. There is no moral grading here. Losing weight is not an achievement of character and needing help is not a failure of one.

In this chapter, you will learn to:

• Take the word "price" apart into the four different numbers it hides
• State a payer's case for refusing to cover obesity therapy at its strongest, and then answer it
• Explain why the shortage happened in a factory rather than in a chemistry lab
• Follow the compounded semaglutide episode end to end and rate the equivalence claim honestly
• Distinguish off-label prescribing from unapproved therapy, and say what it shifts and to whom
• Read the incentive structure of a telehealth platform without assuming anyone is a villain
• Apply Chapter 11's rationing framework and predict a harm that will be large and nearly invisible
• State four positions on obesity fairly, including the ones you disagree with
• Recognize why appetite suppression is dangerous in a population the trials deliberately excluded

Learning Paths

Not optional for any path. This is the one place where the same molecule looks completely different depending on which country you live in and what your employer's plan renewal looks like.

💊 GLP-1 — read every word. §12.1, §12.2 and §12.8 are the chapter; §12.11 is what you will use in a clinic room. 🏋️ Performance — §12.4 to §12.7 generalize far beyond metabolic drugs; they are the machinery of Part III, where the products have no approved version to be compared against at all. 🔬 Science — §12.3 explains why a molecule you can make is not a product you can ship, and §12.9's closing argument is a reasoning error that appears in every field. 💄 Cosmetic — §12.6 and §12.7. The off-label boundary and the bundled telehealth model are the two structures the cosmetic peptide market runs on; Chapter 30 assumes you met them here. 🏥 Clinical — §12.2, §12.6 and §12.10 are practice-relevant. Read §12.9 even if you have already decided, because your patients hold all four positions.

12.1 The price question, stated accurately

Ask "why does this drug cost so much?" and you have already lost, because the sentence contains a word that refers to at least four different numbers. Almost every public argument about GLP-1 pricing is two people using one word for different quantities.

THE FOUR PRICES — one drug, four numbers, none of them the others

  ① LIST PRICE (WAC)      set by the manufacturer. Public. The headline number.
        │                 Nobody with insurance pays it. Some people without insurance do.
        │   ── confidential rebates and fees, manufacturer → benefit manager → plan ──►
        ▼
  ② NET PRICE             what the plan actually pays. Lower than ①, sometimes far lower.
        │                 NOT PUBLIC. Estimated by outsiders, known by insiders.
        │   ── plan design: formulary tier, deductible, coinsurance, quantity limits ──►
        ▼
  ③ OUT-OF-POCKET         what a covered person hands over. Near zero, or most of ①,
                          depending entirely on plan design.

  ④ CASH PRICE            what an uninsured person pays. Historically tracks ① closely,
                          because ① is the only price they have standing to be charged.
                          Manufacturer cash programs are changing this, and changing fast.

  The rebate lowers ② and ③ for people inside a plan. It does nothing for ④.
  That asymmetry was the whole story of Chapter 11's insulin case. It is about to be
  the whole story of this one.

Read that twice, because the rest of the chapter depends on it. The commonest error in GLP-1 cost discourse is quoting a branded list price as the amount society spends per patient. It is not; the net price is lower, and how much lower is a closely held commercial secret, which is itself worth noticing. The second commonest error is the mirror image — observing that net prices are far below list and concluding the affordability complaint is overblown. That answer is unavailable to the uninsured person, who is quoted something near ①.

Why is the list price what it is? Three forces, and cost of manufacture is not among them.

Exclusivity. A patented drug with no generic competitor is priced at what payers will bear, not at cost plus margin. That is not a scandal; it is the design. Patent exclusivity is how societies pay for drug development, and the price during exclusivity is the payment. Whether the payment is proportionate is exactly the argument — but the structure is intentional.

Population size. This is what makes GLP-1 pricing novel. Most expensive drugs are expensive and rare; an orphan therapy at an enormous per-patient price still produces a manageable total. Here a high per-patient price meets an eligible population measured in tens of millions in the United States alone. The unit price is a pharmaceutical problem; the product of price and population is a fiscal one, and they need different arguments.

Geography. Countries that negotiate as a single buyer obtain substantially lower prices for the same molecule in the same pen. This is monopsony, the buyer-side mirror of monopoly, and it works. Whether that means the US subsidizes global innovation or simply fails to negotiate is a live argument. What is not arguable is that the price difference is not a difference in the drug. It is a difference in who is doing the buying.

The part Chapter 7 promised to complicate

Chapter 7 told the incretin story as a triumph, and it was one — roughly forty years of physiology, most of it done in universities and public institutes on public money. It flagged that this chapter would complicate the picture.

The complaint. The public funded the basic science; a private firm patented a molecule built on that foundation; the same public pays again, at a price set by exclusivity, to obtain the result.

The answer. Basic physiology is not a drug. Between "GLP-1 exists and does this" and "here is a stable analog with a workable half-life, a validated process, a safety database in tens of thousands of humans, and a pen a person can use at home" lies an enormous, mostly private effort with a high failure rate — and the failures were paid for out of the revenue of the successes.

Where that leaves you. Both are true, and noticing that beats picking one. The real question is not whether the public contribution deserves recognition but what form it should take: pricing conditions on publicly funded discoveries, public equity stakes, march-in rights, or nothing at all because the present arrangement produces the drugs. Each option has a plausible failure mode.

🔍 Check Your Understanding

1. A news article gives a monthly price for a branded GLP-1 and a commentator replies that "nobody actually pays that." Which of the four prices is each talking about, and which people does each statement describe accurately?
2. Why does the rebate structure lower cost for insured patients while leaving uninsured patients exposed to something near list price?
3. Explain, without using the word "greed," why the same pen costs less in a country with a single national purchaser.

12.2 Why insurers resisted covering obesity

For decades, weight-loss drugs were one of the few categories payers excluded by rule rather than by clinical review. In the United States, the largest public drug program has operated under a longstanding statutory exclusion of agents used for weight loss — written into law, not decided about any particular product. (That exclusion has been under active revision; its current status is exactly the sort of thing to check against a primary source rather than a textbook.) Many commercial plans copied it. Employers bought plans with the category carved out, often without knowing it.

Some of that history was prejudice. But the strongest version of the payer's case is a real argument, and if you cannot state it you cannot answer it.

One: the denominator, not the price. Payers routinely cover drugs costing far more per patient per year. What they have never covered is a drug at this price for a condition affecting this fraction of their membership. Multiply a plausible per-member cost by the share of members meeting eligibility criteria and the result frequently exceeds the plan's entire pharmacy budget. A benefit that cannot be funded without a premium increase is funded by a premium increase, paid by every member including those who will never take the drug. That is a real distributional decision, not an excuse.

Two: chronic therapy means buying it forever. Benefit reverses on discontinuation (§12.9). A payer covering a statin takes on an indefinite obligation too — at a fraction of the cost, for a generic molecule. Here it is large, indefinite, and indexed to a growing eligible population.

Three: the wrong-pockets problem. Members change plans every few years, so a payer investing today to prevent an event fifteen years from now is probably investing on a competitor's behalf. Every payer knows this and discounts long-horizon prevention accordingly — which is why prevention is systematically underfunded in systems that churn.

Four: the field's own history. Fenfluramine combinations were withdrawn after an association with valvular heart disease. Sibutramine was withdrawn after a cardiovascular signal. Rimonabant was withdrawn in Europe and never approved in the US over psychiatric effects. A payer burned three times by drugs that looked good on weight and bad on outcomes learned to wait for hard endpoints.

Five: the category framing. Weight loss was coded as cosmetic, elective, or lifestyle — outside the definition of medical necessity. This part was mostly prejudice, and it interacted badly with the other four: once a category is "lifestyle," the budget argument never has to be made honestly, because nothing is being weighed against anything.

What changed

Efficacy data got large enough to be unignorable. STEP 1 studied semaglutide 2.4 mg weekly in adults with overweight or obesity without diabetes and reported roughly −15% of baseline body weight at 68 weeks against roughly −2.4% on placebo, with both arms receiving lifestyle intervention — which matters, and which §12.6 returns to. STEP 2, in adults with type 2 diabetes, reported roughly −10%. Previous pharmacotherapy did not reach those magnitudes.

But weight was never what payers were waiting for. They were waiting for point four: a hard endpoint.

🔬 Read the Study — SELECT, and why you must always state both numbers

Design. Approximately 17,000 adults with established cardiovascular disease and overweight or obesity, without diabetes, randomized to semaglutide or placebo, followed roughly three years, with a composite of major adverse cardiovascular events as the primary endpoint.

Result. A 20% relative reduction in major adverse cardiovascular events. In absolute terms the event rate fell from roughly 8% to roughly 6.5% — about 1.5 percentage points — corresponding to a number needed to treat of roughly 65 to 70 over that period.

Why both numbers, always. "Twenty percent reduction" and "1.5 percentage points" describe the identical result. The first is what appears in headlines; the second is what a person deciding whether to take a drug for three years actually needs. Neither is dishonest. Quoting only the first is. A relative risk reduction without a baseline risk is uninterpretable — the same 20% is enormous in a high-risk population and nearly worthless in a low-risk one, and SELECT enrolled a deliberately high-risk population.

What it established. That in this specific population, semaglutide reduced hard cardiovascular events. That is a medical claim with a medical endpoint, and it is what moved obesity therapy out of the "lifestyle" bucket in the eyes of payers who had put it there.

What it did not establish. That the same benefit occurs in people without established cardiovascular disease — the great majority of people who want this drug. How much of the benefit is weight loss versus direct vascular effects. Anything about a decade. And it cannot be extrapolated downward by assuming the relative reduction holds: even if it does, absolute benefit shrinks with baseline risk and the number needed to treat grows. Chapter 5's first rule applies — this claim carries a population inside it.

Notice what SELECT did for a payer, which differs from what it did for a patient. It supplied a hard endpoint and a defined subgroup. A plan that could not afford everyone could now cover a population with an outcomes trial behind it and decline the rest on clinical rather than categorical grounds. Evidence does not only expand coverage. It draws lines, and lines have people on both sides.

💊 In the Clinic — what prior authorization is actually doing

Where obesity pharmacotherapy is covered, it is usually covered through a gate: prior authorization, step therapy, documented eligibility criteria, quantity limits, sometimes required enrollment in a lifestyle program. Each has a defensible rationale — checking that the drug is used in the population where it was studied, sequencing cheaper options first, enforcing the label (§12.6).

Each is also, in practice, an attrition device — not a cynical reading but a measurable property. Every administrative step between a person and a prescription loses some fraction of that population, and the fraction lost is not random. It is concentrated among people with less time, less continuity of care, less familiarity with appeals, worse English, and nobody to make phone calls for them. A gate passable in principle by everyone is passed in practice by the people best equipped to pass gates.

Hold both facts. A plan can be doing responsible stewardship and rationing by inconvenience at the same time. The honest question is not whether gates should exist but whether a given gate is calibrated to clinical appropriateness or to attrition — and the test is whether anyone measures how many eligible people it turns away.

12.3 The shortage, and why money could not fix it quickly

For an extended period, demand for GLP-1 receptor agonists exceeded supply, and both manufacturers and the regulator said so publicly. Prescriptions went unfilled. Dose strengths appeared and disappeared. Pharmacies rationed. People with type 2 diabetes — the population several of these products were originally approved for — could not obtain a medication they had taken for years, because demand from the weight indication had consumed the supply.

Why did a company facing the largest demand signal in modern pharmaceutical history not simply make more? The answer is counterintuitive and is the most useful piece of manufacturing knowledge in this book: the binding constraint was aseptic fill-finish capacity and injector-pen assembly — not peptide synthesis.

FROM MOLECULE TO PEN — where the bottleneck actually was

  [1] API SYNTHESIS          make the peptide, attach the fatty acid.
      ▓▓▓░░░░░░░             Chemistry. Scalable with reactors and time. NOT the constraint.

  [2] PURIFY & RELEASE       chromatography, characterization, QC testing. Scalable.
      ▓▓▓▓░░░░░░

  [3] STERILE FORMULATION    clean rooms, validated environments, monitoring. Hard.
      ▓▓▓▓▓▓▓░░░

  [4] ASEPTIC FILL-FINISH    ◄══ THE BINDING CONSTRAINT ══►
      ██████████             Filling sterile liquid into sterile cartridges without
                             contaminating it. Enormous capital cost, years of construction,
                             regulator-inspected LINE BY LINE, PRODUCT BY PRODUCT.

  [5] DEVICE ASSEMBLY        ◄══ THE OTHER BINDING CONSTRAINT ══►
      █████████░             A pen is a precision mechanical device with a dose-setting
                             mechanism. Not chemistry. High-tolerance manufacturing at a
                             scale of hundreds of millions of units.

  [6] PACKAGING & LOGISTICS  cold chain, serialization, distribution. Scalable.
      ▓▓▓░░░░░░░

  Money buys 1, 2 and 6 quickly. Steps 4 and 5 take YEARS, and the years are mostly
  construction, validation and regulatory inspection — none of which compress.

An aseptic filling line is not a machine you order; it is a building. It needs classified clean-room space with validated air handling and personnel-flow design, bespoke equipment, a media-fill validation campaign demonstrating the line produces sterile product repeatedly, and then a filing in which that specific line, at that site, making that product, is reviewed and approved. Regulators do not approve "capacity" in the abstract. They approve a process. Pen assembly is a different problem with the same slowness: a precision mechanical assembly with a dose-setting mechanism that must not fail, built to tolerances closer to consumer electronics than to a vial, at a scale that makes new capacity an industrial project on the order of a car plant.

Three consequences. A capital announcement is not capacity — ask when a line will be validated and approved, not when it will be built; the gap is years. Allocation during a shortage is an ethical decision made privately — whether to protect continuing patients over new starts, whether to prioritize the diabetes indication, which strengths to keep in production. Those are triage judgments with real consequences, made inside companies rather than in public. And the people harmed first were not the people in the headlines: the most sympathetic victims were patients with type 2 diabetes on a stable regimen who lost access. Hold onto that harm — it is concrete and attributable, which will contrast sharply with §12.8.

🧬 The Molecule — the peptide is not the expensive part

Chapter 32 covers manufacturing properly, but one point belongs here because it demolishes a bad argument you will meet constantly: the peptide costs almost nothing to synthesize, therefore the price is pure extraction. The premise is closer to true than most people expect. The conclusion does not follow.

A drug product is not an active ingredient. It is a sterile, stable, correctly concentrated solution, in a container that maintains sterility, delivered by a device a frightened non-professional can operate correctly, made under a validated and inspected process, with a characterized impurity profile and stability data establishing how long it stays itself. The molecule is one line item on that list, and not the largest.

This pays off two sections from now. When a compounded product costs a fraction of the branded one, the saving is not mostly a cheaper molecule — it is everything else on that list. Some of those omissions matter enormously and some barely at all, and distinguishing them is §12.5's whole job. The same reasoning applies with far more force to the Part III gray market, where the only thing purchased is the molecule, and often not even that (Chapters 19 and 34).

12.4 Compounded semaglutide, start to finish

Chapter 6 opened this episode and deliberately declined a verdict. This section tells it end to end. The verdict is still declined — for reasons that should feel earned rather than evasive by the end.

What compounding is. A compounding pharmacy prepares a medication tailored to an individual patient: a child needing a strength nobody manufactures, a patient allergic to a dye in every commercial formulation, someone who cannot swallow a tablet, a discontinued product a few patients still depend on. Old, legitimate, and often essential — not a loophole but a genuine function that industrial manufacturing does not serve.

The general rule. Because compounded products are not reviewed for safety or efficacy before sale, US law generally bars compounders from producing what amounts to a copy of a commercially available approved drug. The approved product carries the entire apparatus of review, and a copy carrying none of it should not compete with it on price.

The exception. That prohibition lifts when the drug is on the regulator's shortage list. A patient who cannot obtain the approved product is not choosing between a reviewed and an unreviewed version; they are choosing between an unreviewed version and nothing. The exception exists so supply failures do not become treatment failures.

What happened. Semaglutide went onto the shortage list and the exception opened. What formed was not a handful of pharmacies serving stranded patients — it was an industry. Telehealth platforms, med spas, weight-loss clinics, and a large number of compounders, both patient-specific and outsourcing-scale, built businesses supplying compounded semaglutide at a fraction of the branded price. Marketing was aggressive and frequently did not mention the product was unapproved. Within a remarkably short period, a very large number of people in the United States were taking a semaglutide product no regulator had reviewed.

Then the shortage resolved. Capacity came online, the regulator removed semaglutide from the shortage list, and with the listing went the exception that had made the market lawful. What followed was messy: litigation challenging the delisting, transition periods of differing lengths for different categories of compounder, wide confusion about what remained permitted, and businesses whose entire model had just been declared to be ending. As of this writing (2026) the aftermath is still working itself out, including questions about which "personalized" versions — different strengths, added ingredients, combinations — are permitted. If you need the current rule, read the regulator's own compounding pages, not a summary. Any book is out of date about this.

⚠️ Hype Check — "it's the same drug for a tenth of the price"

"Compounded semaglutide is the exact same molecule as the brand name. You're just not paying for the marketing."

What's true in it. More than critics usually concede. Semaglutide is a defined molecule, and a facility obtaining genuine semaglutide base and formulating it competently is handling the same active substance the branded product contains. The branded price does include substantial marketing spend. And during a genuine shortage this market delivered therapy to people who could obtain none.

Where it fails. Four places, each checkable. First, "the same molecule" was frequently not true, because of salt forms (§12.5). Second, the delivery device is missing: a pre-filled pen with a dose-setting mechanism is an engineered defense against the commonest error in self-administered medicine, and replacing it with a vial and a syringe reintroduces the error class the device existed to eliminate. Third, "compounded" is not one thing — the category spans a registered, inspected outsourcing facility and an operation whose active ingredient came from a supplier nobody has ever seen. Fourth, the price comparison is against the wrong number: for an insured person the relevant figure is out-of-pocket cost (§12.1, price ③), and for many the compounded product is not cheaper at all.

Verdict: true of a subset of the category, false of the rest, and stated in a form that gives the buyer no way to tell which they have. That is the actual problem — not that compounding is bad, but that the claim erases the variance that matters.

The three users Chapter 6 refused to collapse

The first could not obtain the branded product at all. A genuine shortage, a prescription that could not be filled, a clinician who supported treatment. This person used a legal exception exactly as designed. Whatever the risks of compounded product, the choice was between those risks and no therapy, under conditions the system created.

The second could obtain it but could not afford it. Not a supply barrier — a price barrier. Insured with the category excluded, or uninsured and quoted something near list. This person was routing around a pricing failure rather than a supply failure, and the pricing failure was real. Judging them requires an answer to what they should have done instead, and "gone without a therapy that would have helped them" is an answer — but say it out loud if it is yours.

The third wanted it for appearance, outside any approved criterion. Not an illegitimate desire — people are allowed to want to look different — but a different transaction. The benefit side contains no medical endpoint, and during a shortage the supply consumed came out of a pool that patients with diabetes were being denied.

Chapter 6 refused one verdict and this chapter honors that refusal. Not because judgment is impossible, but because these three share only a product. They differ in what they were choosing between, what alternatives existed, what risk they accepted, and what their choice cost other people. A single verdict necessarily gets at least two of them wrong. If you find yourself with one opinion covering all three, the opinion is about a category rather than about people — and Chapter 5 has a rule for that.

12.5 503A, 503B, and the salt-form problem

"Compounding" covers two legally distinct activities, and the distinction determines almost everything about what a given product is likely to be.

That last row matters most and is read least. Compounded products are not FDA-approved. They are not reviewed for safety, not reviewed for efficacy, and their manufacturing is not reviewed before marketing. A 503B facility is substantially more regulated than a 503A pharmacy — CGMP compliance and federal inspection are meaningful — but inspected and approved are different words for different things. Nobody at a regulatory agency has looked at a compounded semaglutide product and concluded it works and is safe, because that is not what the pathway does.

Regulators raised three specific concerns, and they are three different kinds of problem.

Salt forms. Some compounded products contained semaglutide sodium or semaglutide acetate rather than semaglutide base. These are not the same substance as the semaglutide in approved products, and their safety and effectiveness had not been established. This is not nomenclature: a salt form is a different chemical entity with potentially different solubility, stability, and behavior, and the regulator's position was that a compounder using one was not compounding semaglutide at all. Part of why they appeared is that ingredient-sourcing rules constrain which substances a compounder may lawfully use, and a salt form was, for some suppliers, the version obtainable. The reasoning from "I want to supply semaglutide" to "I am supplying semaglutide sodium" is economic, not pharmacological.

Dosing errors. The approved products come in pre-filled pens with dose-setting mechanisms. Compounded product typically comes as a vial and a syringe. Regulators and poison control centers reported adverse events consistent with dosing errors, including errors arising from confusion between units of measurement when converting between how a dose was expressed and the markings on a syringe. This is precisely the error class a metered device exists to prevent. This book gives no instruction whatsoever on measurement, preparation, or administration, and the reason is visible here: the error mode is real, it has caused harm, and the response is a clinician and a device, not a paragraph.

Provenance. The active ingredient comes from somewhere, and regulators raised concerns about material from facilities that were not registered, not inspected, and sometimes not identifiable. When the supplier of the API cannot be established, the identity, purity, and potency of what is in the vial cannot be established either. Everything downstream of an unverified ingredient is unverified, no matter how competent the pharmacy is. Chapters 19 and 34 cover the analytical reality.

📊 Evidence Rating

The claim: "Compounded semaglutide is equivalent to the branded product."

The rating: ❌ Hype outpaces evidence — for the claim as generally stated. (Rated as of 2026.)

The one-sentence reason: The category spans everything from a registered 503B facility using genuine semaglutide base under inspected quality systems to material of unclear origin supplied as a salt form without the delivery device the approved product was studied with, and no equivalence claim can be true across that range.

What would change it: Product-specific evidence — identity and potency testing establishing semaglutide base, a documented and inspected API supply chain, a validated sterile process, stability data for that formulation, and a delivery method of comparable dosing accuracy. Equivalence claimed for that product, supported by that documentation, is a different and far more defensible claim.

What is conceded, and it is not small: A legitimate 503B facility using genuine semaglutide base under real quality systems produces the same active molecule. Bioequivalence has not been demonstrated and delivery still differs, but the pharmacological premise is sound — and during a genuine shortage this route solved a real problem for real patients with no other access. The ❌ is a rating on the general equivalence claim, not a verdict on every product or every person who used one. Chapter 5's second rule is doing heavy work: a ❌ describes the state of the evidence, not the merit of the molecule and not the character of the patient.

🩺 Safety and Risk — how to think about an unapproved version of an approved drug

When an approved product exists you have an unusually good reference point: known identity, characterized impurities, validated process, delivery device, stability data, safety database. An unapproved version differs from that reference in a specific, enumerable set of ways. Your job is to enumerate them.

• Is it the same substance? Base or salt — and who established that, how?
• Where did the ingredient come from? A registered, inspected supplier, or an unnamed one?
• Under what quality system? 503B under CGMP and federal inspection, 503A under state oversight, or outside both?
• How is it delivered? A metered device, or a measurement the user performs?
• What is the stability? Compounded beyond-use dates are frequently much shorter than an approved product's shelf life, and the difference is not arbitrary.
• Is anything else in it? Added vitamins or other actives take the product further from anything ever studied. "Personalized" is a marketing word, not an evidentiary one.

A competent supplier can answer all six, and the answers are documents rather than reassurances. If nobody will answer, that is the answer. None of the six is about whether semaglutide works — that question is settled and is not what is at issue. Whether any of this is right for you is a conversation with a clinician who knows your history, not a conclusion available from a book.

12.6 Off-label prescribing: what it is, and when it is legitimate

Off-label prescribing means using an approved drug outside the terms of its approval — a different condition, population, dose, or route. In the United States it is legal. Regulators approve products for indications; they do not license the practice of medicine, which is regulated at state level and has always included authority to use an approved drug according to clinical judgment. Manufacturers may not promote off-label uses, and that asymmetry — legal to prescribe, illegal to market — is the load-bearing part of the arrangement.

Off-label is not the same as unapproved. An off-label prescription is for a product that went through full review: identity, manufacturing, and safety profile were examined and accepted. What was not reviewed is the specific use. An unapproved drug is one where none of it was reviewed. Collapsing the two produces bad reasoning in both directions — people who think off-label means dangerous, and people who think "it's just off-label, like half of pediatrics" covers products never approved for anything.

It is common, and frequently the standard of care. Large fractions of pediatric prescribing are off-label, because trials in children are difficult and ethically constrained. Much of oncology operates off-label, because tumor biology crosses the organ-based categories labels are written around. So does much of psychiatry. A categorical ban would be a catastrophe.

What it shifts, and onto whom. When a use is on-label, an expert agency has reviewed a body of evidence and concluded benefit exceeds risk for that population. When a use is off-label, nobody has done that on your behalf. Evidentiary responsibility transfers to the prescriber, who must be able to articulate why this drug, for this patient, for this purpose, is supported — an obligation met with varying seriousness:

The last row is neither hyperbole nor rare. §12.7 is about the structures that produce it.

The label's last clause, and what it is actually telling you

Approved indications for GLP-1 receptor agonists in weight management are narrower than public conversation suggests. They specify a body-mass-index threshold, or a lower threshold in the presence of at least one weight-related comorbidity. (Thresholds vary by jurisdiction and label version, so this book states no numbers; look them up in the current label for the product in your country.) And then they add a clause almost nobody reads:

as an adjunct to a reduced-calorie diet and increased physical activity

That clause is the trial protocol written into the label. It is not a moral instruction and not a hedge. In STEP 1 both arms received a lifestyle intervention, so the roughly 15% versus roughly 2.4% comparison is drug-plus-lifestyle against placebo-plus-lifestyle. A label describes the conditions under which the effect was measured, because that is what a label is for.

Two consequences follow, pointing in opposite directions. First: a person taking the drug with no lifestyle component is not in the trial population, and the honest statement about their expected outcome is that we do not precisely know it. It is almost certainly still substantial — the drug's effect on appetite does not require a coaching program to operate — but that is a different epistemic object from "roughly 15% at 68 weeks," and you should not quote the second for the first situation.

Second, cutting against the moralizing reading: the clause is routinely weaponized, used to imply the drug is a supplement to willpower and that people who do not simultaneously overhaul their lives are misusing it. That is not what a protocol descriptor means. It means the trial had a co-intervention. It does not establish that the co-intervention is necessary, does not establish how much it contributed (the placebo arm's roughly 2.4% is the best hint, and it is small), and does not license anyone to treat the clause as a character test.

🔍 Check Your Understanding

1. A clinician prescribes a GLP-1 agonist to a patient who does not meet the label's threshold. Is that illegal? Is it unapproved? Is it unreasonable? Give three separate answers.
2. Why does off-label prescribing shift evidentiary responsibility, and to whom?
3. State what the "as an adjunct to a reduced-calorie diet" clause does establish, and two things people commonly claim it establishes that it does not.

12.7 Telehealth prescribing and the incentive structure

A large share of GLP-1 prescribing now begins on a website. That deserves even-handed treatment, being neither the scandal nor the liberation it is usually presented as.

What telehealth genuinely solved. Obesity medicine is a small specialty concentrated in cities and academic centers; for many people the nearest clinician with real expertise is hours away and booking months out. Telehealth collapsed that. It also collapsed something less measurable: people with obesity report consistently that medical encounters are places where they are judged, where every complaint is attributed to weight, and where they have been told to lose weight without being offered any means of doing so. Substantial numbers avoid care as a result. A consultation from your kitchen, with no waiting room and no scale in a hallway, removes a barrier invisible to people who have never faced it. That is a real access gain and should not be sneered at.

Now the structure. Many — not all — direct-to-consumer platforms bundle three historically separate things: the clinical consultation, the prescription, and the sale of the product. One company employs or contracts the prescriber, operates the platform, and books revenue when the prescription is written and refilled.

That is a description, not an accusation. It does not imply fraud — many clinicians on these platforms screen properly, decline inappropriate requests, and follow up, and bundled models exist throughout medicine, from the physician who dispenses in the office to the practice that owns its imaging suite. It does imply a direction of pressure: when the entity deciding whether you need a product also sells it, the default resolution of an ambiguous case is predictable — not certain, predictable — and at volume, predictable defaults become the shape of the business.

And it implies checkable design choices, because incentives express themselves in interfaces:

• Does the intake have a failure mode? If the only outcomes are qualifying and not qualifying, with no path to "you should see someone in person first," it is a sorting mechanism, not an assessment.
• Can a person tell which answer qualifies them? Self-reported height and weight are trivially adjustable. If the qualifying answers are obvious, eligibility criteria are advisory — which matters most for §12.10.
• Is there follow-up that could result in stopping? Genuine monitoring includes outcomes that reduce revenue: dose reduction, discontinuation, referral out. And is stopping as easy as starting, or does auto-refill make continuing effortless and discontinuation an act requiring effort?
• What is screened for? Family history of medullary thyroid carcinoma and MEN2, pancreatitis, gallbladder disease, pregnancy plans, current medications, and eating disorder history are all clinically relevant. Whether a given intake asks is a fact you can check in four minutes.

Be fair to the comparison: fee-for-service rewards visits, and a storefront weight-loss clinic selling meal replacements and memberships is at least as conflicted as any website. The point is that you should be able to name the incentive in whatever setting you are in, and naming it does not require believing anyone acts in bad faith.

⚠️ Hype Check — "medically supervised"

"Our program is fully medically supervised — a licensed provider reviews every patient."

What's true. A licensed clinician is almost certainly involved, because a prescription legally requires one. The statement is accurate as far as it goes.

Where it fails. It is unfalsifiable, and it is doing rhetorical work its literal content cannot support. "Medically supervised" has no threshold. It is equally true of a service where a physician spends twenty minutes taking a history and one where a nurse practitioner approves a queue of pre-screened forms. Both are supervision by a licensed provider. They are not the same product.

Ask checkable questions instead. Will I speak to a clinician, synchronously? Can they decline to prescribe, and how often does that happen? Who do I contact at 2 a.m.? Will anyone follow up in a way that could result in stopping? Is my regular clinician informed? Each has a yes-or-no answer, and a genuinely well-supervised service answers five of five without hesitating.

Verdict: not false, but engineered to be unfalsifiable. Chapter 5's fifth rule applies to marketing as well as to evidence — if a claim cannot be wrong, it is not telling you anything.

12.8 Who gets access — and Chapter 11's prediction, answered

Obesity is not randomly distributed. In the United States and most high-income countries its prevalence is patterned by income, occupation, education, geography, food environment, and — in the US context, inseparably from those — race. That patterning is well documented and not seriously disputed. What causes it is disputed, and §12.9 is where that argument lives.

Access is also not randomly distributed, and it runs the other way. It requires insurance with the category covered, or the ability to pay cash indefinitely; a clinician willing to prescribe and able to navigate prior authorization; time for appointments and appeals; a stable address for cold-chain delivery; and increasingly an internet connection and the literacy to use a platform.

Put the two together and you have the defining structural fact of this drug class: the burden is concentrated where access is scarcest, and access is concentrated where the burden is lightest. This is a specific instance of what was described in 1971 as the inverse care law — the availability of good medical care tends to vary inversely with need in the population served. It was not coined about this drug. It fits exactly.

There is a further twist. Within the population that can pay, a meaningful share do not meet approved criteria at all — people using the drug cosmetically, who are wealthier and healthier than average. So during a shortage, supply flows partly to those with least to gain, because they have most capacity to buy. Markets allocate by willingness and ability to pay, and willingness to pay is not a measure of medical need. That is not a failure of markets; it is what markets are. It becomes a policy question only if you think medical need should be the allocation criterion — and if you do, you must say what mechanism would apply it.

Chapter 11's five features

Chapter 11 also observed that ①–③ are properties of the disease and ④–⑤ are properties of a health system. That is why type 1 diabetes is biologically identical in every country while insulin rationing deaths are not: the disease is a constant and the deaths are a variable, which locates the cause in the variable.

Its prediction was that GLP-1 agonists share ③ and ④ but not ①. They share ⑤ too, and arguably more severely — an employer can drop an entire weight-management category at a plan renewal, which happens to whole workforces at once and essentially never happens with insulin. They share ② partially.

What follows from missing ①? This is the crux, and it is not the good news it sounds like.

Because interruption is not acutely lethal, cost-driven discontinuation produces no deaths anyone can attribute, no emergency admissions with an obvious cause, no named victims, and no headlines. The insulin story became politically actionable because it produced funerals with names attached; a young person dead of ketoacidosis while rationing a cheap-to-manufacture drug is a story a legislature can respond to, and price caps followed. That causal chain required the death to be identifiable.

What GLP-1 discontinuation produces instead is this. A person stops. Appetite returns, because it was never gone, only suppressed (§12.9). Weight is regained over months to a couple of years. The cardiovascular risk reduction SELECT measured erodes with it. Three, five, eight years later, some fraction of those people have the infarction or stroke the therapy would have delayed or prevented. Nobody writes "discontinued therapy due to formulary change in 2026" on a death certificate in 2033. The event is separated from its cause by years, mediated by a dozen other factors, spread thinly across an enormous population, and invisible in every dataset anyone currently maintains.

And the population is enormous. Insulin rationing affects a population in the low millions in the US, with a subset rationing at any time. GLP-1 eligibility on approved criteria is measured in tens of millions. If even a modest fraction start, benefit, and then discontinue for cost — and real-world persistence data suggest discontinuation within the first year is common rather than exceptional — the aggregate lost benefit could exceed the total harm of insulin rationing by a substantial multiple.

And it would be almost entirely illegible. One harm is small, concentrated, acute, attributable, and produced legislation. The other is potentially much larger, diffuse, chronic, unattributable, and will produce nothing — because policy responds to legible harm and this harm has no face. There is no advocacy organization for people whose heart attack was moved four years earlier by a coverage decision, because those people do not know that is what happened, and neither does anyone else.

Three caveats, because this is a prediction and not a finding. It is reasoning from structure, and the magnitude is not established; anyone who gives you a number has invented it. It depends on durability of benefit and rate of discontinuation, both being studied and neither settled. And the same illegibility applies to the benefit: nobody will ever point to the heart attacks that did not happen, which is an argument for coverage that is equally hard to make vivid.

What would make it legible? Linked payer-and-outcomes cohort studies following people who discontinue after coverage loss, with hard endpoints and long follow-up. Natural experiments around formulary changes — a large employer dropping coverage is an exogenous shock and a study design. Discontinuation registries. None of it is exotic; it is ordinary pharmacoepidemiology pointed at a question nobody is obliged to answer. Until somebody does it, the honest statement is that we have a strong structural argument that a large harm is occurring and essentially no direct measurement of it — itself exactly the kind of statement this book exists to teach you to make.

12.9 Obesity as disease, behavior, or environment

Underneath everything else sits a question medicine has not resolved: what kind of thing is obesity? Whether the drug should be covered, who should get it, and whether needing it forever is a failure all depend on the answer. There are four serious positions. This book holds none of them. Each is stated as its best advocates would state it, followed by its most serious difficulty.

Position 1 — Obesity is a disease

The case. Body weight is regulated, not chosen. A network involving leptin, ghrelin, insulin, GLP-1, and hypothalamic circuits maintains adiposity around a defended set point and defends it actively. Lose weight and the system responds: hunger rises, satiety falls, and resting energy expenditure declines by more than the lost tissue mass predicts — metabolic adaptation. The body treats weight loss as a threat and mounts a counter-response, measurably, for years. Susceptibility is strongly familial: body-mass index is among the more heritable common human traits in twin and adoption designs, and rare single-gene disorders of the leptin-melanocortin pathway produce severe early-onset obesity that no account of personal choice can accommodate. Obesity meets standard definitions of disease — identifiable pathophysiology, predictable course, characteristic complications, demonstrated response to treatment — and major medical bodies have classified it as one.

The difficulty. Body-mass index is a crude individual marker, so a category defined by it includes many metabolically healthy people and misses some unhealthy ones. Disease classification is partly a social and administrative act determining reimbursement and professional jurisdiction, which means it has interested parties. And the argument from treatment response is a logical error, addressed below.

Position 2 — Obesity is behavior

The case. Energy balance is not optional. Whatever the regulatory system does, adiposity changes only through intake and expenditure; those are the proximate causes and no amount of endocrinology repeals them. And behavior change works: some people achieve substantial, durable loss through sustained changes in eating and activity, and they exist in every population studied. The stronger form is not about blame but about agency. Medicalizing a condition moves the locus of control from the person to the clinician and the product; it teaches, implicitly, that the situation does not respond to your own action — discouraging, and for many people false — and it commits a person to an indefinite pharmaceutical relationship in place of skills that would have been theirs to keep.

The difficulty. Proximate cause and cause are not the same. Every behavior occurs in a body with a particular regulatory system and an environment with particular affordances; the mechanism through which something happens does not identify why it happens. This position must also explain the consistently poor long-term success rate of behavioral weight loss in trials — and the honest explanation, that most attempts are defeated by the counter-regulation of Position 1, concedes a great deal. In practice, whatever its proponents intend, it supplies the vocabulary of weight stigma, which has independent, documented negative health effects.

Position 3 — Obesity is a food environment

The case. Prevalence in most high-income countries rose dramatically over a few decades, in genetically stable populations. Genes do not change in forty years. Environments do. No mechanism exists by which the character of an entire population degrades on a schedule. What did change is enumerable: caloric density and availability, portion norms, the share of diet composed of industrially formulated products engineered for palatability, marketing including to children, the decline of occupational physical activity, built environments that make walking impractical, sleep duration, and the price of calories relative to income. If the cause is structural the remedy is structural — and treating a population-level problem one body at a time, with an indefinite prescription, is expensive, endless, and leaves the generator untouched. Every year the environment produces a new cohort.

The difficulty. Structural change is slow, politically difficult, and has a mixed evidence base. Meanwhile people are sick now, and "the real solution is structural" has been used, not always in good faith, to argue against treating individuals. It also under-explains variation within a shared environment: two people in one household with the same food supply frequently have very different outcomes, and that difference is where Position 1's biology lives.

Position 4 — Health outcomes, not weight, should be the target

The case. Weight is a proxy, and a mediocre one. What we care about is cardiovascular risk, glycemic control, mobility, functional capacity, and mortality — and those improve with physical activity, dietary quality, and sleep partly independently of weight change. Targeting the proxy instead of the outcome is a category error. And weight-focused treatment carries under-counted harms: weight cycling; reinforcement of weight stigma, which independently predicts worse health and care avoidance; precipitation of eating disorders (§12.10); and a clinical history in which weight-loss interventions have repeatedly caused harm. The withdrawn drugs of §12.2 are not ancient history — they are the immediately preceding generation.

The difficulty. The independence of fitness from adiposity is real but partial; adiposity carries risk that fitness attenuates rather than abolishes. And the position is in genuine tension with SELECT, where a weight-targeting drug reduced hard cardiovascular outcomes — precisely the evidence this position asks for. Some advocates have been slow to update. Others have, and their revised view — that weight-focused treatment can be justified by outcomes while weight-focused culture remains harmful — is more coherent and worth taking seriously.

They are not mutually exclusive

These are presented as rivals and are not, in fact, rivals. Two confusions do most of the work.

Heritability is not destiny and does not explain population change. Heritability describes how much of the variation within a population at one time traces to genetic variation. It says nothing about change over time, and a trait can be highly heritable within every generation while its average is driven by environment. Height is the standard case: highly heritable, and average adult height in many countries rose substantially within a century for nutritional reasons. Positions 1 and 3 answer different questions and can both be right.

Proximate mechanism is not ultimate cause. Position 2 is right that intake and expenditure are the channel; Position 3 that the channel's inputs were reshaped; Position 1 that individuals differ enormously in how their regulatory systems respond to the same inputs. A complete account holds all three: an environment that changed, acting through behavior, on a population with biologically variable susceptibility.

And here is the argument this chapter most wants you to keep:

A drug that works on a system is not evidence that the system caused the condition.

Aspirin relieves headache; headaches are not aspirin deficiency. Beta-blockers improve survival in heart failure; heart failure is not adrenaline poisoning. That GLP-1 receptor agonists produce large weight loss establishes that GLP-1 signaling is a powerful lever on appetite and energy balance. It does not establish that obesity is a GLP-1 deficiency, and it does not adjudicate between the four positions at all. A lever is not a diagnosis. This is Chapter 5's third rule running in reverse: there the error was inferring efficacy from mechanism, here it is inferring etiology from efficacy.

Regain is not failure — and the expectation that it is comes from somewhere

When people stop a GLP-1 agonist, weight typically returns; randomized withdrawal designs show substantial regain, and the result is often reported as a scandal.

It is expected physiology, for a knowable reason. These drugs do not replace an absent hormone. There is no GLP-1 deficiency being corrected. The drug applies a sustained pharmacological override to an intact system actively defending a set point. Remove the override and the intact system resumes doing what it was doing throughout. The regain is not the treatment failing; it is the treatment stopping, and the underlying condition still being there.

That makes this chronic therapy, in exactly the sense that antihypertensives, statins, and inhaled corticosteroids are. And here is the observation to carry out of this chapter: nobody says an antihypertensive failed because blood pressure rises when you stop it. No one calls a statin a disappointment because LDL returns to baseline on discontinuation. The expectation that a weight drug should produce a permanent change persisting after it is stopped is applied to no other chronic therapy in medicine.

So where does it come from? From the moral framing. If obesity is fundamentally a defect of character or discipline, then treatment is properly a course of correction — something that fixes the person, after which the corrected person maintains the result under their own power. Under that model, needing the drug indefinitely means the correction did not take. Under the chronic-disease model, needing it indefinitely means the condition is chronic, which is what chronic means.

You can hold any of the four positions and still see this clearly: "it just comes back" is not a scientific finding. It is a moral framework leaking into a clinical judgment.

🔍 Check Your Understanding

1. State the food-environment position in two sentences its strongest advocates would accept, then state its most serious difficulty.
2. Semaglutide produces large weight loss. Using a non-metabolic example, explain why that does not establish that obesity is caused by inadequate GLP-1 signaling.
3. Someone says: "These drugs don't work — as soon as you stop, the weight comes back." Identify the factual claim, agree with it, then explain why the conclusion does not follow.
4. Height is highly heritable, and average height rose substantially within a century. Explain how both are true, and apply the same reasoning to obesity.

12.10 Eating disorders, and a population the trials excluded

This section is written clinically and without moral commentary, because the subject deserves both.

Appetite-suppressing medication interacts badly with restrictive eating disorders, for structural rather than incidental reasons. Restrictive eating disorders require the sufferer to override hunger continuously, and hunger is the principal biological force opposing restriction. A drug that substantially reduces appetite removes the friction restriction has to overcome — and supplies a medical framework that renders the behavior legible as treatment rather than as illness.

Several distinct mechanisms of concern have been described. Facilitation: reduced appetite lowers the effort cost of restriction, potentially allowing intake to fall further and for longer than the person could otherwise sustain. Legitimization: a prescription, a clinician, and a monitored weight trajectory can justify behaviors that would otherwise be recognized as disordered. Masking: signals used to detect a developing problem, such as preoccupation with hunger or distress around eating, may be attenuated when appetite itself is pharmacologically reduced, so deterioration can look calm. Overlap: nausea, vomiting, early satiety, and reduced intake are recognized adverse effects of this drug class and also features of eating disorders, and difficulty of attribution delays recognition.

And the other direction, stated honestly. Binge eating disorder is a different condition with a different relationship to this class. Many people report that GLP-1 agonists reduce intrusive, repetitive thoughts about food — often called "food noise" — and for people whose distress centers on loss-of-control eating, that relief can be significant. Early clinical work here exists and is, as of this writing (2026), preliminary rather than settled: small studies, short durations, unresolved questions about durability and about whether reduced binge frequency reflects genuine improvement or pharmacological suppression of a behavior whose drivers remain. That is a ⚠️-shaped body of evidence at best. The same molecule can help in one eating disorder and harm in another, which is exactly why "is this drug safe in eating disorders?" is a badly formed question.

The screening gap is the practical problem. Most prescribing pathways here were designed around body-mass index, weight-related comorbidities, thyroid and pancreatic history, pregnancy status, and current medications. Eating disorder history is frequently not among them. And the structure of the assessment makes the gap worse: a person with a restrictive eating disorder often presents as an excellent candidate on paper — highly motivated, compliant, focused on weight outcomes, willing to follow instructions precisely.

Detection is a clinician's assessment, not a self-assessment. This is easy to hear as condescension, which it is not. Impaired insight into the severity of one's own eating behavior is a characteristic clinical feature of these disorders — a feature of the illness, not a failing of the person — and it is precisely why self-report screening performs poorly. A book cannot tell you whether you have an eating disorder, and neither can a questionnaire you fill in alone. A conversation with a clinician trained to conduct it can, ideally one not employed by the entity selling the medication.

The evidence base does not cover this population, and you should know why. Trials of GLP-1 agonists for weight management generally excluded people with active eating disorders — a reasonable, ethically standard exclusion, since enrolling them would expose a vulnerable group to unstudied risk. The consequence is often forgotten: the safety data does not extend to the excluded population. There is no reassuring signal because the trials were not looking, and Chapter 5 is explicit about what to do with that — absence of a signal in a population that was excluded is absence of data, not evidence of safety.

One more thing regularly missed. Eating disorders occur at every body size, including in people who meet criteria for obesity, and they are diagnosed later and less often in larger bodies, because clinicians and families read restriction and rapid loss as success rather than symptom. A person in a larger body losing weight rapidly through disordered behavior is frequently congratulated. That is a documented failure of recognition and it belongs in any honest discussion of who this drug class can hurt.

🩺 Safety and Risk — where this section ends

This book does not diagnose, screen, or advise. It can state three things accurately.

One. Eating disorders are treatable, and outcomes are meaningfully better with earlier treatment. They are also serious — anorexia nervosa carries one of the highest mortality rates of any psychiatric illness — and that is a reason to seek assessment, not a reason for alarm.

Two. If you have a history of disordered eating and are considering, or already taking, an appetite-suppressing medication, that history is relevant clinical information and belongs in a conversation with a clinician qualified to assess it. Not because you have done anything wrong, and not because the answer is necessarily no — a history of an eating disorder is not an automatic contraindication, and clinicians manage this. Because the assessment requires someone trained to make it.

Three. If you are worried about someone else, the useful move is not to monitor their eating. It is to help them reach an assessment. Eating disorder helplines and national services exist in most countries and are listed in Appendix M, with the caveat that contact details change and should be verified.

No judgment attaches in either direction. Wanting to weigh less is ordinary. Having a complicated relationship with food is extremely common. Neither tells you what you need, and the person who can tell you is a clinician who has met you.

12.11 What you can do with all of this

1. Take "cost" apart before arguing about it. Identify which of the four numbers in §12.1 a claim refers to and which population it describes accurately. A statement can be true of price ② and false of price ④ while sounding like a statement about "the price."

2. Ask what evidence covers the specific use being proposed for you. Not the molecule — the claim, with its population and endpoint (Chapter 5, rule one). Is this the population that was studied? Is this the endpoint that was measured? If not, what supports the extrapolation? A prescriber practicing well can answer that, and it is not an adversarial question.

3. If a compounded product is on the table, ask the six questions in §12.5. The answers should be documents. If nobody will answer, you have your answer.

4. Have the discontinuation conversation before you start. This is the most neglected conversation in the field and the one most likely to matter to you personally.

💊 In the Clinic — the conversation almost nobody has

The standard pre-treatment discussion covers benefits, side effects, contraindications, and monitoring. It very often does not cover the exit. Questions worth raising before starting:

• If my insurance drops this category at renewal, what happens? Is there an appeal? A manufacturer program? What is the realistic cash cost, and for how long could I sustain it?
• What is the plan if I have to stop — monitoring, an alternative agent, a behavioral program to lean on? What should I expect physically and psychologically, and what is a reason to call?
• Who is managing this in a year — you, or a platform? Does that entity have an interest in whether I continue?
• If this is chronic therapy, is the plan chronic? And if I cannot sustain it, is starting still right for me?

That last question is real, and the answer is frequently still yes — time-limited benefit is still benefit, and delayed events are still delayed. But it is a different decision than most people think they are making, and better made deliberately at the start than discovered at a plan renewal.

5. Read the incentive in whatever setting you are in — not to assume bad faith, but to know which way ambiguity will resolve. Apply it to platforms, clinics, sellers, and whoever funded the study you are reading. Apply it to critics too: a commentator whose audience rewards outrage at pharmaceutical companies has an incentive as legible as a company's.

6. Separate "I disapprove" from "the evidence is weak." These feel identical from the inside and are completely different claims. This is Chapter 5's fourth rule — never downgrade with distaste — and this chapter is the hardest place in the book to obey it.

7. Date your beliefs. Every number here has a shelf life measured in months. As of this writing (2026) is not throat-clearing; it is the epistemic status of the section. Check the regulator's compounding pages, the shortage database, the current label in your jurisdiction, and the trial registry.

8. And decide about the three users yourself. Chapter 6 declined a verdict and this chapter has declined it twice — not because the question is unanswerable, but because you now hold everything required to answer it. If your judgment differs across the three, you have understood the case. If it is identical, check whether you are judging a category rather than three sets of circumstances.

📋 Your Evidence Dossier

Field 10 — Status.

Fields 1 through 9 describe what a molecule is and what the evidence says, and they change slowly, on the timescale of new studies. Field 10 describes what is true about obtaining it, and that changes on the timescale of regulatory decisions, manufacturing capacity, patent expiry, formulary renewals, and litigation. An entry with a current Field 1 and a two-year-old Field 10 will mislead you badly, because it will look authoritative while describing a world that no longer exists.

FIELD 10 — STATUS
  Regulatory status     approved / approved for other indications only / unapproved /
                        compounded under a specific legal pathway / research chemical /
                        withdrawn. State the JURISDICTION — status differs by country.
  Approved indications  what the label actually says, where it matters to you
  Availability          in supply / constrained / on the shortage list / discontinued
  Cost structure        which of the four prices you can find, and which you cannot
  Who is selling it     manufacturer / licensed pharmacy / 503A / 503B / telehealth platform /
                        clinic / online vendor with no verifiable identity
  What is NOT known     the honest gaps
  DATE CHECKED          ◄── mandatory. An undated Field 10 is not information.
  Where to re-check     the specific primary sources for this entry

Worked demonstration — compounded semaglutide

FIELD 10 — COMPOUNDED SEMAGLUTIDE            [worked demonstration]

  Regulatory status   NOT FDA-approved. Compounded products are not reviewed for safety or
                      efficacy. Compounding a copy of an approved drug is generally barred in
                      the US EXCEPT while the drug is on the FDA shortage list. Semaglutide
                      was listed during the shortage and later removed, which narrowed the
                      permission — amid litigation and transition periods of differing lengths
                      for different compounder categories. Unsettled and evolving.

  Approved indications  NONE. A compounded product has no approved indication because it has
                      no approval. The BRANDED products have narrow indications (a BMI
                      threshold, or a lower threshold plus a weight-related comorbidity, "as
                      an adjunct to a reduced-calorie diet and increased physical activity").
                      Those do not transfer.

  Availability        Widely available through telehealth platforms, med spas and clinics
                      during the shortage. Narrowing as of this writing (2026). Any current
                      statement here must be re-verified.

  Cost structure      Marketed at a steep discount to the branded LIST price (price ①).
                      Frequently NOT cheaper than a covered patient's OUT-OF-POCKET cost
                      (price ③). Compare against the right number.

  Who is selling it   Enormously variable, and this is the field that matters most:
                        · registered 503B outsourcing facility, CGMP, FDA-inspected
                        · 503A pharmacy, patient-specific, state-board oversight
                        · telehealth platform that also employs the prescriber (§12.7)
                        · online vendor with no verifiable pharmacy identity
                      These are not the same product and should not share a dossier line.

  What is NOT known   Whether a given vial contains semaglutide BASE or a salt form
                      (sodium/acetate — NOT the same substance; safety and efficacy not
                      established). API provenance. Impurity profile. Actual potency. Real
                      stability under the buyer's storage. Whether anything was added.

  DATE CHECKED        [enter today's date — and mean it]

  Where to re-check   FDA compounding pages and compounding risk alerts; FDA Drug Shortages
                      database; the state board of pharmacy for any 503A; FDA's registered
                      outsourcing facility list for any 503B; DailyMed for the branded label.

The same field for the branded product would read: approved, with narrow indications that differ by product and country; availability constrained and then improving, verify current status; list price public, net price confidential, out-of-pocket entirely dependent on plan design; sold by one identifiable manufacturer through licensed distribution plus emerging cash channels; not known — net price, decade-scale outcomes, and effects in excluded populations including active eating disorders.

Notice what the two entries do side by side. They differ in one field far more than any other: who is selling it — for one, a single identifiable regulated entity; for the other, a category spanning inspected facilities and anonymous vendors. That variance is the whole reason the equivalence claim earned a ❌ in §12.5, and Field 10 is where it becomes visible. Do this field now for each of your dossier peptides. For the Part III compounds you will find Field 10 is the entire story, and that "who is selling it" has no good answer at all.

Conclusion

The evidence for GLP-1 receptor agonists in weight management is among the strongest in this book. The questions in this chapter are among the least settled, and that combination is the point.

List price, net price, an insured person's copay, and an uninsured person's cash price are four different numbers, and most public argument is two people using one word for different quantities. The shortage was not a chemistry problem — it was aseptic fill-finish and injector-pen assembly. The compounded market formed inside a legal exception built for exactly the situation that had arisen, grew far beyond what the exception contemplated, and narrowed when the shortage resolved. Off-label prescribing is legal, frequently necessary, and shifts the evidentiary burden onto the prescriber. Bundled telehealth genuinely expanded access and genuinely aligned revenue with prescribing, both at once.

Chapter 11 asked this chapter to answer a prediction, and the answer is not comforting. GLP-1 agonists share insulin's inelastic demand, its residual-payer structure, and its discontinuous coverage — but not its acute lethality on interruption. That missing feature is why cost-driven discontinuation here will produce no funerals, no named victims, and no legislation, while plausibly generating an aggregate loss of benefit across tens of millions of people that exceeds insulin rationing several times over. A harm that large and that invisible is a policy failure waiting for a study design.

Underneath it sits a question medicine has not resolved, and four serious positions on it, none excluded by the others. This book picks none. What it insists on is one logical point that survives whichever you choose: a drug working on a system is not evidence that the system caused the condition. And regain after discontinuation is expected physiology rather than treatment failure, because the drug overrides an intact system rather than replacing a missing one — an expectation applied to no other chronic therapy, and one that comes from a moral framework rather than a scientific one. You are allowed to hold moral views about all of this. You are not allowed — not if you want to reason well — to let them change what the evidence says.

Chapter 13 returns to the molecules. Tirzepatide adds a second receptor, retatrutide a third, and the question becomes whether more targets means more benefit or simply more surface area for things to go wrong. It should not feel like an escape: everything here applies to those molecules too, and sooner than anyone expects.

Key Terms

List price (WAC) — the price a manufacturer publishes. Public, paid by almost nobody with insurance, and the number that appears in headlines.

Net price — what a plan actually pays after confidential rebates and fees. Not public.

Rebate — a payment from manufacturer to plan or benefit manager in exchange for formulary placement. Lowers net price for insurers; does nothing for the uninsured.

Pharmacy benefit manager (PBM) — an intermediary that negotiates drug prices and manages formularies for health plans.

Formulary — the list of drugs a plan covers and on what terms.

Prior authorization — approval a prescriber must obtain before a plan will pay. A stewardship tool that also functions as an attrition filter. Step therapy is its sibling: a cheaper option must be tried and fail first.

Out-of-pocket cost — what a covered patient actually pays at the counter, set by plan design rather than by the drug's price.

Residual payer — the party left holding the undiscounted price because they sit outside the rebate structure. In US drug pricing, typically the uninsured patient.

Monopsony — a market with a single dominant buyer; the buyer-side mirror of monopoly.

Fill-finish — the stage in which sterile drug product is filled into its final container under aseptic processing. Capital-intensive, heavily regulated, slow to expand, and the binding constraint in the GLP-1 shortage.

Drug shortage list — a regulator-maintained public list of drugs in shortage. Listing opens the exception permitting compounded copies of an approved drug.

Compounding — pharmacy preparation of a medication tailored to a patient. Legitimate and often necessary; compounded products are not FDA-approved and are not reviewed for safety or efficacy.

503A pharmacy — a traditional compounding pharmacy preparing patient-specific prescriptions, regulated primarily by state boards.

503B outsourcing facility — a compounder registered with FDA that may produce larger batches without patient-specific prescriptions, must comply with CGMP, and is FDA-inspected. Still not approved.

Salt form — a drug substance paired with a counter-ion. Semaglutide sodium and semaglutide acetate are not the same substance as semaglutide base, and their safety and effectiveness had not been established.

Active pharmaceutical ingredient (API) — the drug substance before formulation. Its provenance determines whether anything downstream can be trusted.

Off-label use — prescribing an approved drug outside its approved indication, population, dose, or route. Legal in the US, common, often the standard of care, and it shifts evidentiary responsibility to the prescriber.

Direct-to-consumer telehealth — a model bundling consultation, prescription, and product sale inside one commercial entity.

Inverse care law — the observation, described in 1971, that the availability of good medical care tends to vary inversely with need in the population served.

Set point — the body weight a regulatory system actively defends through changes in hunger, satiety, and energy expenditure. Metabolic adaptation is part of that defense: energy expenditure falls after weight loss by more than the lost tissue mass predicts.

Chronic therapy — treatment whose benefit depends on continued administration and reverses on discontinuation.

Relative risk reduction — the proportional reduction in event rate between arms. Uninterpretable without the baseline risk. Absolute risk reduction is the difference in percentage points, and number needed to treat is its reciprocal — how many people must be treated, for how long, to prevent one event.

Spaced Review

1. (Ch. 5 + Ch. 12) Someone says "compounded semaglutide is ❌, so it doesn't work." Identify two distinct errors using Chapter 5's frozen rules, then state the ❌ claim from §12.5 in its correct form, with population and endpoint attached.

2. (Ch. 8 + Ch. 12) Chapter 8 explained how the substitution at position 8 and the fatty-acid attachment give semaglutide its long half-life. Relate those design choices to §12.3's claim that the bottleneck was fill-finish and pen assembly rather than synthesis — and explain why a weekly injectable creates a device-manufacturing problem that a daily tablet would not.

3. (Ch. 6 + Ch. 12) Chapter 6 introduced three people who used compounded semaglutide: one who could not obtain the branded product during a genuine shortage, one who could obtain it but not afford it, and one who wanted it cosmetically outside approved criteria. Give a separate judgment for each, and for each state explicitly what alternative you are holding them to. If your three judgments are identical, say why the differences do not matter.

4. (Ch. 11 + Ch. 12) Name which of Chapter 11's five features GLP-1 agonists share and which they do not, and explain why the missing feature makes the resulting harm larger in aggregate and smaller in political visibility. Then name one study design that would make it measurable.

5. (Ch. 12) A relative is starting a GLP-1 agonist through a telehealth platform and asks what to check. Give five questions with checkable answers, and say for each what a bad answer looks like. Include nothing about dose, preparation, or technique.