Before the Conclusion | After Approval: Medicines Reconsidered, case 3
If a medicine genuinely relieves pain and causes less gastrointestinal damage than an older drug, it is natural to see that as progress. But if that progress also comes with more heart attacks and strokes, the original assessment must be recalculated.
Vioxx, or rofecoxib, once occupied exactly that position. It entered the US market in 1999 for conditions including osteoarthritis and for pain treatment. On September 30, 2004, the company announced its worldwide withdrawal. The immediate trigger was a long-term randomized trial that revealed serious cardiovascular risks.[1,2]
Its story shows that withdrawing a drug does not necessarily mean it never had a therapeutic effect. It can also mean that the benefits originally measured covered only part of the patient's overall situation.
Why people wanted a new class of pain medicines
Nonsteroidal anti-inflammatory drugs such as ibuprofen and naproxen reduce pain and inflammation, but can also damage the gastrointestinal tract, causing ulcers and bleeding in severe cases. For arthritis patients who need long-term medication, this is a real burden.
Rofecoxib is a selective COX-2 inhibitor. The development rationale was to improve the safety of pain treatment by more selectively inhibiting a process involved in inflammation while preserving some protective functions of the stomach lining. This selectivity had a biological rationale and a target that could be tested clinically.[3]
Yet a target in the human body usually participates in more than one function. COX-2-related pathways also contribute to the production of substances such as prostacyclin, which are involved in vascular function and platelet regulation. Early human pharmacology research had already suggested that selective COX-2 inhibition might affect that balance.[4]
Such mechanistic evidence can generate a risk hypothesis, but cannot independently tell us how many additional heart attacks will occur. Ultimately, more complete patient outcomes still need to be compared in clinical data.
One trial, results pointing in two directions
The VIGOR study enrolled 8,076 patients with rheumatoid arthritis, comparing 50 mg of rofecoxib daily with 500 mg of naproxen twice daily. Its primary aim was to compare gastrointestinal safety.[3]
The rofecoxib group had fewer serious gastrointestinal events, supporting one of the advantages originally hoped for. But the same study also produced a signal in the opposite direction: more heart attacks in the rofecoxib group.
This raised a crucial interpretive problem. Was the difference mainly due to rofecoxib increasing risk, or to some cardiovascular protective effect of naproxen? Both could also be operating at once. Because the comparator drug had biological effects of its own, that single comparison could not easily disentangle the two contributions.
This was more than a statistical dispute. If the difference mainly reflected protection from the comparator, physicians might retain confidence in rofecoxib. If the new drug itself increased serious harm, its use required greater caution.
But the existence of another possible explanation cannot make an unfavorable signal disappear. The reasonable next steps were to investigate the risk, fully disclose existing information, and reassess conditions of use as evidence accumulated. Testing an explanation required new comparisons.
A placebo control makes the risk harder to dismiss
APPROVe was originally designed to investigate whether rofecoxib could reduce recurrence of colorectal adenomas. It enrolled 2,586 patients with a history of adenomas, randomly assigning them to 25 mg of rofecoxib daily or placebo, with longer-term observation.[5]
Confirmed thrombotic events occurred in 46 patients receiving rofecoxib and 26 receiving placebo. Accounting for follow-up time, event rates were 1.50 and 0.78 per 100 patient-years, respectively, with a relative risk of approximately 1.92. The excess events chiefly included heart attacks and ischemic cerebrovascular events.[5]
A relative risk close to two does not mean every patient's risk rose by nearly 100 percentage points. We must also consider absolute event rates, patients' baseline risks, and duration of exposure. Even when an individual's probability of an event is not very high, additional harm can become substantial when a medicine is used widely and for long periods.
The placebo comparison meant that “naproxen is simply protective” could not independently explain all the findings. It also used a lower rofecoxib dose than VIGOR, so the earlier unfavorable result could not simply be attributed to that trial's higher dose.
In 2004, the company announced withdrawal on the basis of APPROVe. Notably, that trial's original purpose did not concern routine joint pain, but the serious safety signal it uncovered could change the assessment of the drug's broader use.[2]
How was information handled after the risk signal appeared?
Describing Vioxx only as “a drug whose previously unknown side effect was discovered after approval” would omit the part of this history that most needs examining. Cardiovascular signals had already appeared before the 2004 withdrawal. How did the relevant institutions and the company investigate, interpret, and communicate them?
In 2005, the New England Journal of Medicine issued an expression of concern about the VIGOR paper. It stated that the material submitted to the journal had omitted three heart attacks in the rofecoxib group, which the editors believed caused the paper to underestimate the risk difference between the groups. The authors subsequently responded, and the editors reaffirmed their concerns in 2006.[6]
These public disputes concern data cutoffs, completeness of reporting, and who knew what and when. They require discussion based on specific documents; every disagreement cannot simply be described as the same finding of unlawful conduct. The chronology must also be accurate: the journal's expression of concern came after withdrawal. The immediate evidence triggering the 2004 withdrawal came from APPROVe.
Nevertheless, information handling itself has clinical consequences. If physicians can clearly see gastrointestinal benefits but not cardiovascular risks with equal clarity, their prescribing choices will be affected. Drug safety depends both on whether studies detect harm and on whether the information enters medical decisions fully and promptly afterward.
Why we cannot ask only “Which has fewer side effects?”
Rofecoxib reminds us that a drug being safer for one organ does not mean it is safer for the patient overall. Reducing gastrointestinal bleeding has value; increasing heart attacks and strokes has a serious cost. Both must enter the same assessment.
That assessment also depends on alternatives. For a rapidly fatal disease with no other options, people may accept substantial risk. For pain with other treatment choices, the same serious risk requires stronger additional benefits to justify it. The comparison concerns the specific options available to the patient.
Risks and benefits are therefore not two fixed labels on a medicine's package. They change with dose, treatment duration, baseline patient risk, and the comparator. A worthwhile use in one setting cannot automatically be generalized to long-term treatment or wider populations.
After Vioxx, the evaluation of this type of treatment extended beyond a single product. In 2005, the FDA comprehensively reviewed cardiovascular data for selective and nonselective nonsteroidal anti-inflammatory drugs, required stronger warnings in prescription labeling, and advanced further assessment of risks across the class.[7]
This historical fact should not be stretched into a claim that “all pain medicines have exactly the same risk.” What it supports is the need to test other important harms even when a drug is newer, more selective, or gentler on the stomach.
A drug's efficacy must be placed within patients' overall outcomes
Feeling less pain after taking medicine is a real and important benefit. Serious cardiovascular harm does not become less meaningful because pain improves. The difficult question is how to assess both together, rather than selecting one as the complete answer.
In my judgment, the most important demand left by Vioxx is that safety signals receive the same serious attention as efficacy results. A trial's primary aim may concern the gastrointestinal tract, but unfavorable cardiovascular findings still warrant full investigation. A mechanism may explain an advantage, but must remain open to challenge from other mechanisms or clinical outcomes.
New evidence changes judgments, and the handling of evidence already available changes patients' lives too. Assessing whether withdrawal was timely requires considering both questions together.
After leaving the market, Vioxx was also the subject of large-scale personal-injury claims and settlements, with their own participation criteria, review, and payment processes. Stopping future sales did not automatically address losses patients had already suffered. We will examine that aftermath alongside other drug cases in the closing article, “After a Drug Is Withdrawn, Who Bears Patients' Losses?”[8]
Next, we turn to bevacizumab. It remains in use for certain cancers but lost its US breast cancer indication. The central dispute there was how to weigh the patient benefit represented by delaying disease progression.
Main sources
[1] Merck, September 30, 2004: withdrawal announcement and information on the market launch date.
[2] EMA, October 6, 2004: statement on Vioxx's worldwide withdrawal and the APPROVe risk.
[3] Bombardier et al., New England Journal of Medicine, 2000: VIGOR randomized trial. The paper later received an expression of concern and should be read alongside [6].
[4] McAdam et al., PNAS, 1999: selective COX-2 inhibition and human prostacyclin production. This human pharmacology experiment studied celecoxib. It illustrates a pathway-level risk hypothesis and cannot be treated as a direct estimate of rofecoxib's clinical risk.
[5] Bresalier et al., New England Journal of Medicine, 2005: APPROVe cardiovascular outcomes. This article uses the event rates and relative risk reported in that original study. It does not interpret time intervals within a single study as personal safety boundaries.
[6] Editors of the New England Journal of Medicine: 2005 expression of concern about VIGOR and 2006 reaffirmation.
[7] FDA, April 6, 2005: review and regulatory recommendations on cardiovascular risks of nonsteroidal anti-inflammatory drugs. This describes a historical regulatory decision and does not replace current labeling for an individual medicine.
[8] US District Court for the Eastern District of Louisiana: Vioxx product-liability litigation and 2007 settlement records. The scope and progress of compensation will be discussed in the closing article.
Sources checked through October 28, 2026. This article is for medical and scientific information exchange and does not constitute individual medical advice. Discuss specific treatment changes with your physician.
Beyond the conclusion, there are questions worth continuing to ask.
Medicine · Science · Ideas and Reading
Follow Before the Conclusion as we stay curious and think independently together.
Text and image usage statement
The text was originally written by Before the Conclusion. You are welcome to share the article's web link. Without separate permission, please do not reproduce or excerpt the text by copying, reposting, or publishing screenshots. Unless otherwise indicated, illustrations may not be separately republished, copied, modified, or used commercially. Rights in cited material remain with their respective rights holders.