The Vioxx Saga: Perspective on the Recall

by William R. Ware, PhD

Merck was granted FDA approval to market the painkiller Vioxx (rofecoxib), a so-called COX-2 inhibitor, on May 21, 1999. The same year Pharmacia (now merged with Pfizer) launched Celebrex (celecoxib), a competing drug. "Me Too" drugs followed, some still in the pharmaceutical pipeline. In the five years that elapsed before Merck withdrew Vioxx from the market an estimated 80 million patients took the drug (1). Vioxx was a "blockbuster" with sales over 2 billion US dollars per year. Editorials in both medical journals and the press have suggested that the FDA and Merck knew that Vioxx had serious adverse effects and were both at fault for not acting earlier, implying that the system protecting the consumer is not working very well. What happened during those five years is thus of considerable interest.

Early warnings of trouble
The first study hinting at trouble (VIGOR) was published in November 2000 (2). Between January and July 1999, over 8000 patients were enrolled in Merck's "Phase 4" post-approval study of upper gastrointestinal (UGI) toxicity of Vioxx. It was a double blind, randomized trial comparing the occurrence of gastrointestinal toxicity of Vioxx and naproxen (Aleve, another non-steroidal anti-inflammatory) in patients with rheumatoid arthritis. The paper indicated that the risk of "important" UGI adverse events was significantly lower in the Vioxx group compared to the naproxen group. As regards the risk of heart attack (MI), the authors claimed that Vioxx significantly increased MI only in people with a previous cardiovascular problem who should have been taking aspirin but did not because of the study design. The hypothesis was also promoted that naproxen was cardio-protective, a hypothesis supported by a Merck review published in 2001 of 16 studies involved in their pre-approval process which found no excess of cardiovascular (CV) adverse events except when Vioxx was compared with naproxen (3).

The November 2000 paper, however, was not based on the complete set of data. Subsequently Mukherjee, Nissen and Topol from the Department of Cardiovascular Medicine of The Cleveland Clinic reexamined the cardiovascular (CV) risk question using all the data submitted to the FDA and the results of the FDA's adjudicated examination of the observed CV events. The results appeared in the Journal of the American Medical Association (JAMA) in August 2001 (4). Based on adjudicated cases, the risk of developing a CV event was 2.38 higher in the Vioxx group compared to the naproxen controls with a probability that this result occurred by a chance of two in a thousand (p=0.002) i.e. a significant result. For those who should have been taking aspirin, i.e. those with history of stroke, transient ischemic attack, MI, unstable or stable angina, bypass surgery or angioplasty, the Vioxx group had a statistically significant relative risk (RR) of 4.89. There is also the question of corticosteroid use (5). In VIGOR over 50% of the subjects were using oral steroids for osteoarthritis or menstrual or other pain, which is not in keeping with the typical patient profile of many users of Vioxx. The authors of the 2001 JAMA article raised serious questions regarding the safety of Vioxx and called for studies to address the question. However, in 2001 the issue was seriously confused by the possibility that naproxen, the control in the VIGOR study, was cardio-protective, an argument Merck also used to explain the enhanced risks seen in their pre-approval clinical trials (3) as well as in the VIGOR study. In April of 2002 the FDA instructed Merck to include precautions about CV risk in the package insert for Vioxx. However, it is well known that many doctors and patients do not study these inserts and that they are a very inefficient method of disseminating a warning (6,7).

In 2002 at least six studies addressed the naproxen hypothesis (the assumption that the adverse CV effects observed with Vioxx when compared to naproxen were due to naproxen being protective rather than to Vioxx being detrimental) (8,9,10,11). No protective effect was seen in two, protection in three, and statistically non-significant protection in one. A clinical study which did not depend on naproxen for a control was published in 2002 in The Lancet (8). It was found that Vioxx doses greater than 25 mg/d gave an increased risk (RR 1.7-1.93). This study had a rather short average period of follow-up, and there was still an issue with corticosteroid use. In 2003 a Canadian study found no increase in CV adverse events with Vioxx and no decrease with naproxen, but this was a very short-term study (average follow-up 144 days) (12). Thus at the end of 2003, the situation did not appear to have been significantly clarified. In spite of at least seven studies, the naproxen hypothesis was still alive, and there was only one additional study which indicated potential CV risk for the users of Vioxx and one that was negative.

Confirmation of Vioxx's adverse effects
In 2004 a study in which naproxen was not an issue was published in the journal Circulation (13). There was no exclusion for a history of CV disease risk. It was found that for Vioxx doses equal to or greater than 25 mg/d, there was an increased risk for acute heart attacks (AMIs) which disappeared after 90 days of use. Then in August 2004, a paper at an International Pharmacoepidemiology meeting in Bordeaux, France reported the results of an FDA sponsored study involving a huge Kaiser Permanente (US Health Maintenance Organization) database which compared the risk of adverse CV events for users of Vioxx vs. Celebrex. The study, which attempted to project the excess of adverse CV events associated with Vioxx, concluded that more than 27,000 MI and sudden cardiac deaths occurred between 1999 and 2003 which could have been avoided by taking Celebrex rather than Vioxx. Taking 25 mg/d or more of Vioxx resulted in over three times the risk of AMI or sudden cardiac death as compared to Celebrex use. The relative risk was 1.5 for lower doses of Vioxx. This alarming result, presented at a poster session, is probably not widely known, although on September 27, 2004 the American College of Rheumatology (ACR) issued a hotline prompted by what they called "substantial media coverage" of this study. However, the ACR did not recommend discontinuing Vioxx, but merely suggested that more studies were needed and clinicians needed to be aware of potential adverse events. A fairly complete account of this FDA sponsored study was "published" after the recall in The Wall Street Journal, October 6, 2004. The full report has been submitted to The Lancet and a version is now available at www.fda.gov. It is important, incidentally, that there is also a supporting biological mechanistic hypothesis that would explain the increase of adverse CV events due to COX-2 inhibition (14). This mechanism applies to all COX-2 inhibitors and has been the basis of calls for more extensive and powerful studies of all drugs in this class.

The fall-out
Editorial comments quickly appeared in The New England Journal of Medicine (1,15) and The Lancet (16) and additional commentary appeared in major newspapers. Eric Topol, one of the authors of the 2001 JAMA paper that reevaluated the VIGOR study, contributed an article to The New York Times (17) titled "Good Riddance to a Bad Drug". Concern was evident regarding the impact of the recall on the drug safety review process, the time required for drug companies to get regulatory approval, and the confidence of the general public in "the system". Over three years elapsed between the publication in JAMA of the full analysis of the VIGOR study and the recall, a recall actually precipitated by a study with an endpoint totally unrelated to the original indications for which Vioxx was marketed.

In the NEJM editorial, Eric Topol suggests that Merck should have initiated or been forced by the FDA to initiate an appropriate trial directed at the question of CV risk, given the growing body of evidence (1). In addition he feels that the uncertain CV risk picture should have prompted Merck to stop the intensive direct-to-consumer advertising (e.g. TV ads) until the matter was resolved. However, he does give the FDA credit for sponsoring the Kaiser Permanente study. In a second editorial comment in the same issue, G. A. FitzGerald takes a similar position (15). In his opinion, "the rational basis for addressing the cardiovascular effects of these drugs (COX-2 inhibitors) has been clear for the past five years, yet even the most fundamental questions have not been addressed directly". He concludes "the burden of proof now rests with those who claim that this is a problem for rofecoxib (Vioxx) alone and does not extend to the other coxhibs. We must remember that the absence of evidence is not evidence of absence". John Abramson, who teaches clinical medicine at Harvard, points out in the recent book Overdo$ed America (5), that patients taking Vioxx had 21% more "serious adverse events" of all kinds (CV, upper GI, etc.) than those on naproxen. He comments "Something is very wrong with a system that leads patients to demand and doctors to prescribe a drug that provides no better pain relief and causes significantly more side effects". This statement was based on data known to the FDA in 2001 and reviewed in theJAMA article of August 2001, as well as on his own reevaluation of the data.

The ugly details
In early November 2004 significant new information became available that helps clarify some of the questions raised by the Vioxx saga. On November 5 a new meta-analysis (where a number of studies are lumped together and re-analyzed) of a large number of Vioxx risk studies was published in The Lancet (18). The Swiss authors had no connection with Merck. At the end of 2000, they found on the basis of 13 studies (20,742 patients) that the RR of an MI due to taking Vioxx was 2.30 (p=0.01) and a year later on the basis of 16 studies (21,432 patients) the RR was 2.24 (p=0.007). They also found that the small protection observed for naproxen was insufficient to explain the increased number of MIs. Thus, based on their view of what was known to Merck and the FDA by the end of either 2001 or 2002, the authors concluded that Vioxx should have been withdrawn. One might ask why the FDA failed to do a meta-analysis in 2001 of this same data rather than leave it up to Merck which had an obvious conflict of interest (3).<

The Wall Street Journal on November 1, 2004 announced that they had come into possession of both emails and internal documents originating from Merck which suggested that management had been aware from before 2000 that there was a problem but publicly downplayed its importance and taught their sales reps how to field awkward questions from doctors regarding CV risks - the printed policy was characterized by the word "DODGE". In an email on March 9, 2000, Merck's research chief told colleagues that the CV events "are clearly there" and called it a "shame" but the company's public statements continued to reject the link between Vioxx and increased risk. The WSJ article also outlines several incidents where Merck attempted to intimidate academic researchers who were making anti-Merck statements. Merck is reported to have taken the position that these emails are being viewed out of context. They also claim that the Swiss meta-analysis had design flaws as did the FDA sponsored study (19).

On November 4, 2004 The New York Times published an article discussing the attempts of the FDA to delay and downplay the results of their study first reported in Bordeaux, France. Dr. David Graham, the lead investigator in the study, provided the newspaper with emails to back up the claim that his effort to publish the study was delayed and demeaned by to FDA officials. In his editorial in NEJM, Eric Topol called for a congressional investigation of the way in which the FDA handled the Vioxx matter (1). Recent reports in the press suggest this may well happen.

Finally, a paper in the December 2004 issue of Atherosclerosis provides a new piece to the puzzle regarding what might be a mechanism for enhanced adverse CV events (20). This paper examined the effect of several COX-2 inhibitors on the susceptibility of human low-density lipoprotein (LDL) cholesterol to oxidative modification. They found that the so-called Sulfone COX-2 inhibitors (Vioxx and etoricoxib) clearly increased the susceptibility of biological lipids to oxidative damage through non-enzyme processes and that this could provide "mechanistic insight" into reported differences in COX-2 inhibitors in connection with adverse CV events since no such increase in susceptibility was found for celecoxib (Celebrex), valdecoxib or meloxicam nor the non-selective inhibitors such as ibuprofen, naproxen and diclofenac. Incidentally, etoricoxib is a Merck drug currently up for approval. The FDA has just announced that it will not approve it before more studies are conducted regarding safety issues. The study in Atherosclerosis should reinforce their decision. This new mechanistic hypothesis is distinctly different from that mentioned above (14), which involved adverse effects of COX-2 inhibitors on the blood clotting process with all coxib drugs implicated.

Flaws in the system
The Vioxx fiasco might easily cause the public to lose confidence in both the FDA, the drug companies and clinical and observational studies, especially when it came so soon after the hormone replacement therapy (HRT) flip-flop (21). After all, taking prescription drugs represents an act of faith. i.e. the implicit assumption that the drug company in question is not concealing serious risks for their own financial gain, and that the regulatory agencies are on guard and seeing to the safety issue with all due diligence. In the absence of such faith, taking prescription drugs is like a crap-shoot. But even if this faith is misplaced, without prescription drugs many areas of modern medicine would be essentially paralyzed. Nor would any reasonable person suggest abandoning medical studies with human subjects just because they sometimes yield misleading, inconsistent or even wrong results, or results that may in some cases be influenced by the source of funding. These studies of necessity employ imperfect tools to investigate what are in many cases exceedingly complex questions involving diverse populations, studies that must rely on statistics to a much greater extent than, for example, studies in the physical sciences or even laboratory studies in the biological sciences. The HRT flip-flop has in fact stimulated considerable discussion of the strengths and weaknesses of observational vs. randomized clinical trials (21,22,23) and improvements are being suggested which will probably benefit the science of epidemiology in the future.

One must not lose sight of the fact that until late 2004 the picture was still rather confused, with inconsistent results both of the Vioxx CV risk and the naproxen hypothesis. Contrary to the position of a number of vocal critics, from what has been outlined above, this does not appear to be an open and shut case. One of the authors of the now famous JAMA article of 2001, Dr. Steven Nissen, is quoted in an interview with Business Week, October 1, 2004, that "Even in August, 2001, when we published our first report on the risks, it was highly speculative. We didn't have hard data. We had some soft data and some suspicions about the mechanism of action. We could just as easily been wrong. When asked if the drug approval process should be changed to avoid such problems in the future, he responded "There's potentially a solution, but it's very costly. We need a more robust post-approval surveillance system to keep track of adverse events with new drugs. The system is voluntary on the part of the drug companies now, but I would have it mandatory. Studies have found that only between 1% and 10% of serious adverse events with drugs on the market are actually reported. So the FDA is making decisions on inadequate data now. This (would be) a very important initiative, in my opinion".

It is probably not generally realized that the number of subjects involved in the trials required for drug approval is small (typically a few thousand, sometimes considerably less), and side effects occurring at a rate of less than 1% are easily attributed to chance since the number of events is very low. Even phase 4 studies generally do not involve really large groups of participants (6,21). Thus the real study of adverse events is highly informal, haphazard and uncontrolled and occurs subsequent to approval, since now the patient base becomes huge as a drug is aggressively promoted both to doctors and directly to the public and free samples flow freely. But, as Nissen points out, the reporting system for adverse drug events is inefficient and only infrequently used, which means that most adverse reactions go unrecorded, a problem which leads to a significant lag time before a serious side effect is recognized, investigated and if necessary, regulatory action taken or "voluntary" recall is seen as the only option by the company involved (6,21,24). Also adverse side reactions may simply go undetected since in many cases the association is difficult or even impossible to make in the normal clinical setting. If an adverse event occurs, let's assume, at a rate of even one per hundred users, how is a physician to make the association with a specific drug if his patient load is say 600 individuals, only a fraction take the drug and similar symptoms are seen in patients not on the drug in question? The only hope would be if the side effect was very unusual. In his editorial Eric Topol estimates that on the basis of total patients exposed, the risk observed in the Merck colon cancer prevention study and VIGOR can be extrapolated to the prediction that there could have been as many as 160,000 MI or stroke incidents related to the total Vioxx exposure (1). Yet post-approval surveillance results appear never to have been cited publicly as a significant source of concern in connection with Vioxx.

Thus the consumer of pharmaceuticals needs to be keenly aware that regulatory approval is not a blanket guarantee of safety for the simple reason that such a guarantee appears in general impossible, and the history of recalls shows that the drug companies understandably are generally in a denial mode until the evidence is overwhelming. The case of Vioxx illustrates the complexity of the problem. Consumers of prescription drugs would be well advised to educate themselves as to the side effects, something that can be done, albeit very imperfectly, by combining information from the pharmacy, the package insert, and the internet or The Physicians Desk Reference if available. However, most drugs have long lists of side effects and it is sometimes difficult to identify which are of critical importance and then ascertain the related symptoms, and some warning signs can only be identified with lab tests. Early action is frequently important for the avoidance of permanent damage, disability or even death. The statin class of cholesterol lowering drugs is a good example where early detection of muscle problems is important if very serious consequences are to be avoided (6). But in the case of Vioxx, it may in many cases have been impossible to detect early warning signs of CV problems related to the drug, especially in the case of individuals with pre-existing CV conditions, and sudden cardiac death, by definition, provides no early warning. Also, it appears unrealistic to expect the regulatory agencies to be in there at the first sign of trouble and take rapid, definitive action the public will hear about on the evening news or read in the newspaper (21). The bottom line is always the matter of risk vs. benefit. But as we have seen in the Vioxx fiasco, this relationship is frequently difficult or even impossible to evaluate.