When the U.S. Food and Drug Administration asked American Home Products to pull fenfluramine (Pondimin) and dexfenfluramine (Redux) from the market on 15 September 1997, it ended a weight-loss phenomenon that had been built on an off-label pairing the agency had never approved and a long-term safety record nobody had assembled. Fen-Phen — the colloquial yoking of fenfluramine, a serotonin-releasing anorectic approved in 1973, with phentermine, a stimulant approved in 1959 — had become, by 1996, one of the most prescribed drug ideas in America: roughly 18 million prescriptions in that year alone and an estimated 6 million users, exposure on the order of 61 million patient-months. The gap between the promise of easy, doctor-blessed thinness and the harm was a class of damage that does not announce itself: scarred heart valves and, more rarely, lethal pulmonary hypertension.
The combination’s popularity rested on a 1992 study and on a regulatory permission slip that was never granted. The FDA had approved fenfluramine and phentermine each as standalone short-term anorectics; it had never approved their combination, nor either drug for the open-ended, cosmetic, multi-year use that fen-phen clinics dispensed. The 1996 approval of dexfenfluramine — the more potent right-handed isomer, sold as Redux and the first new prescription weight-loss drug in 23 years — poured accelerant on the market, despite an FDA advisory committee that had initially voted against it and known pulmonary-hypertension concerns in European data.
The verdict is plain at the outset. A combination the agency never sanctioned, marketed for a use it never authorized, reached millions before any long-term cardiac safety study existed; the damage was found not by the regulator or the manufacturer but by Mayo Clinic cardiologists who noticed a pattern. Their report, published in the New England Journal of Medicine on 28 August 1997, described 24 women with no prior cardiac history who had developed an unusual valvular disease — leaflets thickened with a glistening, carcinoid-like plaque — after taking fen-phen. The drug came off the market eighteen days later.
What followed was, for its era, the largest product-liability reckoning in pharmaceutical history. American Home Products (renamed Wyeth in 2002) agreed to a $3.75 billion national class settlement in 1999, then watched its total liability climb past $21 billion as tens of thousands of plaintiffs opted out, and the case became the standard byword for what happens when an off-label combination and a direct-to-consumer obesity market outrun the safety science that should have preceded them.
When Bayer AG voluntarily withdrew Baycol (cerivastatin) from the worldwide market on 8 August 2001, the company described it as a precaution against a rising “reporting rate” of muscle injury; the documented record shows that the latecomer statin had already been associated with fatal rhabdomyolysis at roughly forty to fifty times the rate of its competitors, a hazard that was foreseeable from its known pharmacology and visible in adverse-event data well before the recall. Cerivastatin, the sixth statin to reach the U.S. market and approved by the Food and Drug Administration in June 1997, was promoted as a potent, low-dose entrant in an already crowded class. It captured slightly under 4 percent of the U.S. statin market — and yet, by the FDA’s own count at withdrawal, it accounted for 31 American deaths from rhabdomyolysis, a condition in which skeletal muscle disintegrates, floods the bloodstream with myoglobin, and shuts down the kidneys.
The gap between promise and reality was not a matter of rare bad luck. In an analysis by FDA scientists, the relative reporting rate of fatal rhabdomyolysis was on the order of forty times higher for cerivastatin than for the other approved statins; for all rhabdomyolysis, fatal and nonfatal, the figure was roughly fifty-four times higher. The danger was concentrated and predictable: it rose with the 0.8 mg high dose Bayer had introduced to compete on potency, rose in elderly patients, and rose catastrophically when cerivastatin was combined with gemfibrozil, a fibrate that interfered with the drug’s clearance. Twelve of the 31 U.S. deaths involved that very combination — a combination Bayer had contraindicated on the label, but which physicians prescribed in numbers the warning failed to suppress.
The withdrawal came only when the body count made the signal undeniable. The verdict here is therefore plain at the outset: an approved cholesterol drug holding a minority share of its market produced a disproportionate share of its class’s fatal harm, because a known drug-interaction hazard and an aggressive high-dose strategy were allowed to reach roughly six million patients while the contraindication label substituted for the market action the data warranted.
What followed was a transatlantic litigation reckoning. Bayer faced on the order of 7,800 product-liability suits in the United States alone — a figure that grew past 12,000 — and ultimately paid well over $1 billion to settle claims, while internal documents pried loose in discovery showed company scientists had flagged the gemfibrozil danger and the rising death reports before the recall. Baycol became the standard cautionary case for how a “me-too” drug competing on potency can convert a known interaction into mass casualties.
When the U.S. Food and Drug Administration asked Warner-Lambert to pull Rezulin from the market on 21 March 2000, it ended a three-year arc in which a celebrated first-in-class diabetes pill killed at least 63 people through liver failure while the warning that should have stopped it had been visible from before launch. Rezulin (troglitazone), the first thiazolidinedione — an oral “insulin sensitizer” acting on the PPAR-gamma receptor — was approved on 29 January 1997 and marketed as a breakthrough that let type 2 diabetics use their own insulin more effectively. The gap between that promise and the harm was not subtle: the drug produced idiosyncratic, sometimes fulminant hepatotoxicity, and Britain’s Glaxo Wellcome withdrew it from the U.K. market on 1 December 1997 — barely ten months after the U.S. approval — while the FDA kept it on American shelves for another twenty-eight months.
The signal was institutional, not merely clinical. The FDA medical officer originally assigned to the application, Dr. John Gueriguian, had recommended against approval over cardiac and liver concerns; he was removed from the review on 4 November 1996, and his negative analysis was purged from the official record before the drug was cleared on an expedited timeline. Within the first year of marketing, serious hepatotoxicity reports accumulated, and an internal FDA memorandum cited 135 reports of serious liver injury by late 1997. Public Citizen’s Health Research Group, led by Dr. Sidney Wolfe, petitioned for withdrawal in July 1998, asking the agency how many more Americans would have to die or need transplants.
The verdict is therefore plain at the outset: a profitable, regulator-blessed, heavily prescribed medicine reached nearly two million Americans while the evidence that condemned it — a foreign withdrawal, a buried internal review, mounting death reports, and an outside petition — sat fully legible in the record. The FDA’s response was incrementalism: four label changes between 1997 and 1999 adding ever-stricter liver-monitoring instructions that real-world prescribing could not reliably execute, rather than removal.
What finally forced the withdrawal was arithmetic the labels could not fix. Once two safer thiazolidinediones — rosiglitazone and pioglitazone — reached the market in 1999 without the same fatal liver signal, Rezulin’s risk-benefit case collapsed, and the FDA requested its removal on 21 March 2000. Litigation followed: Warner-Lambert, absorbed by Pfizer in 2000, ultimately resolved roughly 35,000 claims for an estimated $750 million, and “Rezulin” became a byword for how a buried safety review and an ignored foreign recall can keep a lethal drug on the market for years.
When Hoechst Marion Roussel pulled Seldane from the U.S. market in 1997 and the Food and Drug Administration formally withdrew its approval effective 4 November 1998, the official posture was prudent stewardship; the documented record shows the company removing a drug it had known to be conditionally lethal for at least six years, and doing so only once it had a patented, profitable substitute ready to sell. Seldane (terfenadine), synthesized by Richardson-Merrell chemists in 1973 and marketed in the United States in 1985 as the world’s first nonsedating antihistamine, was a genuine therapeutic advance: it relieved hay fever without the drowsiness that made older antihistamines hazardous to drivers and workers. That advance concealed a defect in the molecule itself.
The gap between promise and harm lay in pharmacokinetics. Terfenadine as swallowed was a prodrug, almost entirely converted by the liver enzyme CYP3A4 into an active metabolite that did the antihistamine work. The unconverted parent compound, however, blocked the hERG (Kv11.1) cardiac potassium channel, prolonging the QT interval and inviting torsades de pointes, a chaotic ventricular rhythm that can kill within minutes. So long as metabolism was brisk, parent levels stayed trivial and the heart was safe. But any common CYP3A4 inhibitor — the antifungals ketoconazole and itraconazole, the macrolide antibiotics erythromycin and clarithromycin, even a glass of grapefruit juice — could throttle that conversion, let the parent drug pile up, and convert an allergy pill into an arrhythmogen. A patient stable for years could be killed by a course of antibiotics for a sinus infection.
The verdict is therefore plain at the outset: the danger was not a late surprise but a measured, published mechanism. FDA reports flagged ventricular arrhythmias by June 1990; a boxed warning followed in July 1992; and the landmark Honig study in JAMA in March 1993 demonstrated, in healthy volunteers, that ketoconazole co-administration made parent terfenadine accumulate and the QT interval lengthen. For most of the drug’s life the response was a warning label on a product that remained on pharmacy shelves and, for a time, was even pursued for over-the-counter status.
What finally ended Seldane was not the body count but the chemistry of replacement. The active metabolite — fexofenadine — carried the antihistamine benefit without the parent compound’s hERG liability. Marion Merrell Dow’s successor patented it, the FDA approved it as Allegra in July 1996, and only then did the maker withdraw Seldane and the FDA move to revoke approval. Terfenadine became the first blockbuster pulled for QT-related arrhythmia, the case that taught regulators to screen new drugs against the hERG channel, and a byword for a harm that was understood, labeled, and tolerated until a safer money-maker was ready.
When Janssen Pharmaceutica — the Belgian unit of Johnson & Johnson — announced on 23 March 2000 that it would stop general U.S. marketing of Propulsid, it presented the move as a precaution; the documented record shows a drug the company’s own labels had flagged as potentially lethal since 1995, kept on the open market for five more years while prescriptions ran past thirty million. Propulsid (cisapride) was a gastrointestinal prokinetic approved by the U.S. Food and Drug Administration in 1993 for nighttime heartburn from gastroesophageal reflux. It worked by speeding the gut, but it also blocked a cardiac potassium channel, prolonging the QT interval and, in vulnerable patients, triggering torsades de pointes — a chaotic ventricular arrhythmia that can stop the heart.
The gap between the indication and the harm was the whole tragedy. Cisapride was approved for adult reflux, but its danger concentrated in people who should never have received it: patients taking common interacting drugs — the macrolide antibiotics erythromycin and clarithromycin, the antifungals ketoconazole, itraconazole and fluconazole — that raised cisapride’s blood levels, and patients with underlying cardiac or metabolic risk. By the FDA’s accounting, of the 341 serious arrhythmias and 80 deaths reported through December 1999, roughly 85 percent occurred in patients with a recognized contraindication or risk factor. The drug was, in effect, safe in the population that did not need protecting and dangerous in the population that did.
The verdict is therefore plain at the outset: an approved, widely prescribed heartburn pill killed through interactions its manufacturer and the FDA had named in the label years before the withdrawal, and a meaningful share of the dead were infants and children who received it off-label for colic and reflux even though it was never approved for pediatric use. One observational study of roughly 58,000 premature infants found that about a fifth had been given cisapride.
What followed was a slow regulatory unwind and a mass-tort settlement. Janssen halted general sale effective 14 July 2000, retaining only a tightly controlled limited-access protocol; in 2004 Johnson & Johnson agreed to pay up to $90 million to resolve claims that the drug caused some 300 deaths and nearly 16,000 injuries. Propulsid became the textbook case of a “Dear Doctor” letter and a black box that did not change prescribing fast enough to stop the dying.
When Hoffmann-La Roche pulled Posicor from the U.S. market on 8 June 1998, just under a year after the Food and Drug Administration approved it in June 1997, the company framed the move as a precaution against drug interactions; the documented record shows that the molecule’s defining pharmacology — potent, mechanism-based inhibition of the liver enzyme cytochrome P450 3A4 — made it a hazard the moment it was co-prescribed with the ordinary medicines a hypertension or angina patient already took. Posicor (mibefradil), marketed as a first-in-class blocker of low-voltage T-type calcium channels, was promoted as a cleaner, more selective antihypertensive. It instead became one of the fastest major drug withdrawals of its era, undone not by a flaw in what it did to the heart but by what it did to the metabolism of dozens of unrelated drugs.
The gap between promise and harm was structural, not incidental. By slowing CYP3A4 — the enzyme that clears a large fraction of all prescription drugs — mibefradil let co-administered medicines accumulate to toxic levels. Statins such as simvastatin built up until muscle dissolved into the bloodstream (rhabdomyolysis), threatening the kidneys; antiarrhythmics, certain antihistamines, calcium-channel blockers and beta-blockers stacked into bradycardia, shock, and cardiac arrest. By the time of withdrawal more than 25 medications were known to be dangerous in combination with Posicor, and the drug was being taken by almost 200,000 Americans and nearly twice that number worldwide.
The verdict here is therefore plain at the outset. This was not a slow-burn safety signal suppressed over years, as with some withdrawals; it was a foreseeable consequence of the drug’s own measured pharmacokinetics, surfacing within months of launch. A particularly damning cluster, reported in JAMA in 1998, described patients who — following the very label advice to switch off Posicor onto another agent — went into shock within twelve hours, one of whom died, because mibefradil’s enzyme inhibition persisted for days after the last dose.
The reckoning was swift rather than litigious. Roche withdrew the drug globally, the FDA documented the action in the Federal Register, and the episode became a standard teaching case in clinical pharmacology: the canonical example of how a single potent CYP3A4 inhibitor can convert routine co-prescription into mass interaction risk, and how a list of contraindicated drugs that keeps growing after launch is itself the warning.
When Wyeth-Ayerst Laboratories pulled Duract from American pharmacies on 22 June 1998, the company cast it as a precaution against rare liver events in patients who had overstayed a clearly printed 10-day limit; the documented record shows the hepatic hazard was visible inside the drug’s own approval file nearly a year earlier, and that an FDA medical officer had argued — and lost — for a boxed warning before a single prescription was written. Duract (bromfenac sodium), a short-term non-steroidal anti-inflammatory approved by the U.S. Food and Drug Administration on 15 July 1997 for acute pain lasting ten days or less, instead became one of the fastest market withdrawals of the modern era. In roughly eleven months it generated some 2.5 million prescriptions and was associated with at least four deaths and eight liver transplants from fulminant hepatic failure, plus a further dozen serious liver injuries, with the count of severe drug-induced liver injury cases ultimately climbing past fifty.
The gap between promise and harm was not a post-market surprise. Bromfenac’s submission already showed that roughly 15 percent of short-term trial patients developed elevations of the liver enzymes AST or ALT to three times the upper limit of normal — a signal of hepatocellular stress markedly greater than that seen with comparable NSAIDs. The FDA medical reviewer who read that data advocated a black-box warning as a condition of approval and was overruled; the drug shipped instead with a routine label and a duration cap of ten days, on the theory that limiting exposure would contain the risk.
That theory collapsed against ordinary prescribing behavior. Acute pain frequently does not resolve in ten days, refills were written, and patients took bromfenac for weeks. The cases that killed and transplanted clustered almost entirely in this extended-use population — the precise scenario the 10-day rule was meant to prevent but had no mechanism to enforce. The verdict here is therefore plain at the outset: an approved, branded analgesic reached millions while the hepatotoxic signal that would condemn it sat fully legible in the agency’s own pre-approval review.
What followed was a rapid, near-total revocation. Wyeth strengthened the warnings in February 1998, the harm continued, and the company withdrew the drug in June; the FDA later formally rescinded the New Drug Application. The molecule survived only by abandoning the bloodstream — reformulated years later as a topical eye drop, where systemic exposure was negligible — while “Duract” became a byword for a withdrawal whose justification had been written before the launch.
When the U.S. Food and Drug Administration announced on 29 March 2007 that the makers of pergolide had agreed to pull the drug from the American market, the action was filed as a routine post-market safety measure; the documented record shows that the fibrotic heart-valve mechanism that condemned Permax had been understood, in chemical detail, for years before the echocardiograms were read. Pergolide (marketed by Eli Lilly as Permax and approved by the FDA in 1988 as an adjunct to levodopa for Parkinson’s disease) was an ergot-derived dopamine agonist that relieved tremor and rigidity by stimulating dopamine receptors — but it also bound potently to the 5-HT2B serotonin receptor on heart-valve tissue, the identical target that had driven the valvulopathy behind the 1997 fen-phen catastrophe.
The gap between promise and harm was not an unforeseeable side effect. Two studies published side by side in The New England Journal of Medicine on 4 January 2007 closed the case. René Schade and colleagues, mining the U.K. General Practice Research Database cohort of 11,417 patients, found that pergolide carried an adjusted incidence-rate ratio of 7.1 (95% CI 2.3-22.3) for newly diagnosed cardiac-valve regurgitation. Renzo Zanettini’s echocardiographic study found clinically important (moderate-to-severe) regurgitation in 23.4 percent of pergolide patients versus 5.6 percent of controls — and, decisively, 0 percent in patients on non-ergot dopamine agonists, which do not hit the 5-HT2B receptor.
The verdict is therefore plain at the outset: a drug with no proven efficacy advantage over safer, mechanistically distinct alternatives kept hundreds of thousands of Parkinson’s patients exposed to a known fibrotic hazard years after the signal was legible. The FDA asked Lilly to add valvulopathy to the warnings section in 2003 and mandated a boxed warning in 2006; neither step removed the drug. Only the 2007 NEJM pair forced a withdrawal the pharmacology had argued for since at least 2002 — a quieter case than the blockbuster scandals, with no settlement fund and no Senate hearing, only a back-of-the-shelf agent kept in front-line use long after a receptor profile shared with a recalled diet drug had told clinicians exactly where to look.