The Vioxx Recall: What Merck Knew Before Pulling the Drug

On 30 September 2004 the pharmaceutical company Merck did something companies almost never do to a drug earning them two and a half billion dollars a year: it withdrew it, worldwide, voluntarily, in a single announcement. The drug was Vioxx, a painkiller taken by an estimated eighty million people since its launch in 1999. Merck’s own long-term trial, running to see whether the drug could prevent bowel polyps, had instead thrown up a result the company could not explain away — patients on Vioxx were suffering heart attacks and strokes at roughly twice the rate of those on placebo. The withdrawal was presented as an act of corporate responsibility, a company doing the right thing the moment the evidence became undeniable. The more uncomfortable question, which years of litigation would work over in detail, was how long the evidence had been becoming undeniable before that moment arrived. This is the story of a signal, and of the distance between when a company sees a signal and when it acts on one.

The promise of a cleaner painkiller

To understand the appeal of Vioxx you have to understand the problem it claimed to solve. The old anti-inflammatory painkillers — ibuprofen, naproxen, aspirin and their kin — are genuinely useful for arthritis and chronic pain, but they carry a real cost: over time they erode the stomach lining and cause ulcers and bleeding, which kill a meaningful number of people every year. Pharmacologists had worked out that these drugs block two related enzymes, COX-1 and COX-2, and that the pain relief came mostly from blocking COX-2 while much of the stomach damage came from blocking COX-1.

The idea, elegant on paper, was to build a drug that blocked only COX-2 — all the pain relief, none of the gastric harm. Merck’s version was rofecoxib, sold as Vioxx; Pfizer had a rival, celecoxib, sold as Celebrex. This new class, the coxibs, launched into an enormous market with an enormous marketing push, and Vioxx became one of the fastest-selling drugs in history. Millions of people with aching joints were switched to it on the promise that it was gentler on the gut.

What the elegant theory had not fully reckoned with was that COX-2 does other things in the body, including in the lining of blood vessels, where its products help keep platelets from clumping. Tip that balance and you might, in theory, make clots — heart attacks and strokes — more likely. The theoretical risk was discussed in the pharmacology before the drug was ever a scandal. The question was whether it showed up in real patients.

The trial that flashed red

It showed up early. In 2000, barely a year after launch, Merck published the results of a large trial called VIGOR, designed to prove Vioxx’s central selling point — that it caused fewer stomach ulcers and bleeds than naproxen, an older painkiller. On that measure the trial succeeded. But it also found something the marketing did not dwell on: patients taking Vioxx had around five times the rate of heart attacks compared with those on naproxen.

Here the story reaches its first genuine fork, and it is a subtle one that mattered enormously. There were two ways to read that gap. Either Vioxx was raising the risk of heart attacks, or naproxen was lowering it — protecting the heart the way low-dose aspirin does. Merck advanced the second explanation, the “naproxen is cardioprotective” hypothesis, and for a while it was not unreasonable on its face, because there was no placebo group in VIGOR to settle it. But the cardioprotective effect of naproxen strong enough to explain the whole gap was never well established, and the alternative reading — that Vioxx itself was dangerous — was left hanging. The company, understandably invested in a blockbuster, leaned hard on the interpretation that spared the drug.

The regulators were not oblivious. The US Food and Drug Administration reviewed VIGOR and, in 2002, required Merck to add information about the cardiovascular findings to the Vioxx label. But a warning buried in a label is a quiet thing, and the drug kept selling in vast quantities while the debate continued in the medical literature. Cardiologists began to raise alarms. In 2001 a group including the cardiologist Eric Topol published an analysis in a major journal arguing that the available data pointed to a real cardiovascular hazard and calling for a proper trial to settle it. The signal was in the open. The drug stayed on the market for three more years.

What the documents showed

The withdrawal in 2004 came from the APPROVe trial, the bowel-polyp study, which had a placebo group and therefore could not be waved away with the naproxen argument. It showed roughly a doubling of heart attacks and strokes in Vioxx patients compared with placebo, and the excess appeared after around eighteen months of use. Merck pulled the drug.

Then came the litigation, and with it the kernel of documented fact that this case rests on. Thousands of plaintiffs sued, claiming Vioxx had caused their heart attacks, and discovery pulled Merck’s internal records into court. Some of what emerged was genuinely damning. Internal communications from the late 1990s, before the drug even launched, showed scientists inside the company already worried about the cardiovascular risk and discussing how it might play out. One internal email, later widely quoted, discussed the difficulty of designing studies in a way that would not surface the very risk they feared; another set of documents suggested the company had, at points, sought to identify and manage physicians who were critical of the drug. A particularly bruising episode concerned the reporting of the VIGOR data itself: an analysis published in the New England Journal of Medicine had, the journal’s editors later charged in an unusual public “expression of concern”, omitted several heart attacks that had occurred in the Vioxx group before the paper went to press, making the drug look safer than the full data supported.

None of this proves that Merck launched a drug it knew to be a killer. What it proves is more textured and, in its way, more instructive: that the company saw the cardiovascular signal early, had internal reason to take it seriously, and consistently resolved the ambiguity in the direction that protected its blockbuster — in how it interpreted VIGOR, in how it presented the data, in how it responded to critics — during the years the drug remained on the market and the exposure kept growing.

Counting the cost, and where the counting gets loose

The human toll became one of the most cited and most slippery numbers in modern pharmaceutical history, and it is worth handling honestly, because the fork here is a fork in the arithmetic.

The most influential estimate came from David Graham, a safety reviewer inside the FDA itself, who used a large database of patient records to estimate the excess cardiovascular events attributable to Vioxx. Testifying to the US Senate in November 2004, he put the likely range at somewhere between 88,000 and 139,000 excess heart attacks in the United States, of which perhaps 30 to 40 per cent were fatal. Those are the figures that entered public memory, often stripped of their range and their caveats, so that “Vioxx killed tens of thousands” became a settled fact.

The estimate is serious and came from a credible insider who took real professional risk to make it. It is also, by its nature, a modelled extrapolation — built on assumptions about background rates, dose, duration and how many of the eighty million users took the drug long enough to matter, since the excess risk in the trials appeared mainly after many months of use. Different assumptions yield different totals, and the honest range is wide. The harm was real and large; what deserves attention is how a carefully bounded estimate hardens, in retelling, into a precise and confident figure. The precision is borrowed; the harm is genuine.

Merck fought the individual lawsuits case by case, won a fair number of them, and eventually in 2007 agreed to a settlement of around 4.85 billion dollars to resolve the bulk of the US personal-injury claims — without, in the manner of these things, admitting liability. The case-by-case strategy was itself instructive. Rather than settle en masse, Merck contested each claim on its individual facts — Did this particular plaintiff take the drug long enough? Did they have other risk factors for a heart attack? — and the strategy worked well enough in the courtroom that the company won early verdicts. It was a reminder that a modelled population-level harm, however real in aggregate, is hard to pin to any single body, and that a well-funded defendant can turn that difficulty to its advantage one juror at a time.

The distance between a signal and an act

What lingers about Vioxx is not a cartoon of corporate evil, because the record does not quite support the cartoon. Merck did, in the end, withdraw a hugely profitable drug on its own trial data, which many companies would have resisted longer. The disturbing part is subtler and more general: the years between the first clear signal in VIGOR in 2000 and the withdrawal in 2004, during which the drug kept selling and the company kept finding reasons to read the danger as something other than what it was.

That gap is where the real lesson lives, and it is a lesson about incentives rather than villainy. A company sitting on a two-and-a-half-billion-dollar product is not a neutral reader of ambiguous safety data. Every genuinely uncertain finding — and VIGOR was genuinely uncertain, which is exactly why it was so dangerous — will tend to be resolved in the direction that protects the product. No single wicked decision is required; a hundred small, defensible, self-interested judgements do the work, each one reasonable in isolation. The same structural pressure runs through the opioid marketing story, where a company read every ambiguity about addiction in the direction that sold more pills, and through the tobacco documents, where an industry manufactured doubt out of genuine scientific uncertainty for decades. It echoes, too, the older pattern of thalidomide: a warning visible before the withdrawal, and a company that treated the warning as a public-relations problem first.

The reason people reach for the word “cover-up” is that the outcome looks exactly as if the risk had been hidden — tens of thousands harmed while the danger was known to some. But the mechanism was not a locked vault of guilty knowledge. It was something more ordinary and more resistant to reform: an organisation that could see the red light and kept finding honest-seeming reasons not to stop, because stopping cost two and a half billion dollars a year. Understanding that is more useful than a verdict, because the vault can be broken open once and the incentive cannot. It is still there, in every company sitting on a blockbuster and reading an ambiguous trial.