|Home | About | Journals | Submit | Contact Us | Français|
In September 2004, rofecoxib was voluntarily withdrawn from the worldwide market. Our objective was to determine whether and when analysis of published and unpublished placebo-controlled trials could have revealed cardiovascular risk associated with rofecoxib before its withdrawal as an example to inform future post-market pharmaceutical safety surveillance efforts.
We conducted a cumulative subject-level pooled analysis of data from all randomized, placebo-controlled trials of rofecoxib conducted by the manufacturer before September 2004. Our main outcome measurement was incidence of any investigator-reported death from any cause or cardiovascular thromboembolic (CVT) adverse event.
We identified 30 randomized, placebo-controlled trials of rofecoxib that enrolled 20,152 subjects. Trial duration ranged from 4 weeks to 4 years, enrollment ranged from 17 to 2586 subjects prescribed either rofecoxib or placebo, and rofecoxib dosage ranged from 12.5 mg to 50 mg. As of December 2000, 21 (70%) of these trials had been completed and the risk of CVT adverse event or death was greater among subjects assigned to rofecoxib, with the difference being borderline statistically significant (Rate Ratio [RR]=2.18, 95% Confidence Interval [CI], 0.93–5.81; p=0.07). Subsequently collected data strengthened the statistical association (as of June 2001: RR=1.35, 95% CI, 1.00–1.96; p=0.05; as of April 2002: RR=1.39, 95% CI, 1.07–1.80; p=0.02).
Cumulative pooled analysis of all randomized, placebo-controlled trials demonstrates a progressing trend toward increased cardiovascular risk associated with rofecoxib compared with placebo as early as December 2000, reaching a P value of 0.05 by June 2001, nearly 3 and a half years before the manufacturer’s voluntary market withdrawal.
The brief appearance on the worldwide market of rofecoxib, a selective cyclooxygenase (COX)-2–selective agent, was a striking commercial success, reaching $2 billion in annual sales soon after its introduction in May 1999. Merck, the maker of rofecoxib, promoted it as a safer alternative to traditional non-steroidal anti-inflammatory drugs (NSAIDs), although there were concerns about its cardiovascular adverse effects early in its development.1, 2 In September 2004, the manufacturer voluntarily withdrew rofecoxib from the market after an interim safety analysis indicated that the drug was associated with increased risk of cardiovascular events within the Adenomatous Polyp PRevention On Vioxx (APPROVe) trial.3 The APPROVe trial, which tested the hypothesis that rofecoxib reduced the risk of colon polyp recurrence (and thus colorectal cancer),3 was terminated early by its Data Safety and Monitoring Board (DSMB). In November 2004, the manufacturer’s Chief Executive Officer (CEO) testified to the United States Senate Committee that “Until data from [APPROVe] …, the combined data from randomized controlled clinical trials showed no difference in confirmed cardiovascular event rates between Vioxx and placebo.”4
Today, the medical community acknowledges the cardiovascular risks associated with rofecoxib.5–7 A Scientific Statement by the American Heart Association affirmed that selective COX-2 inhibitors have adverse cardiovascular effects, including increased risk for myocardial infarction, stroke, heart failure, and hypertension.8 A pertinent question is whether the Food and Drug Administration (FDA) and the manufacturer could have known before the time of its withdrawal that rofecoxib increased the risk of cardiovascular events, when compared with placebo, through the analysis of the manufacturer’s published and unpublished clinical trial data. Three company-sponsored and conducted meta-analyses demonstrated no increased cardiovascular risk associated with the drug while it remained on the market, in 2001,9 2002,10 and 2003,11 although none of the three focused on placebo-controlled studies or used all available data.
This question is particularly relevant as we consider current and future challenges in drug safety monitoring. Can the example of the rofecoxib experience inform and improve the approach used by industry and regulators to conduct post-market pharmaceutical safety surveillance? If all available data, both published and unpublished, had been independently and iteratively analyzed, could the increased cardiovascular risk have been known earlier? Understanding whether this approach may be informative is of even greater importance after the September 2007 enactment of the FDA Amendments Act (FDAAA) in the United States, which not only requires the sponsors of all drug, biologic, and device trials to register their studies, at inception, in the publicly-available ClinicalTrials.gov database (with the exception of Phase I clinical trials), but also to update the registry for approved drugs and devices within 12 months of study completion (24 months if the studied drug is currently under review at the FDA) to include trial results: primary and principal secondary outcomes as well as safety outcomes. Therefore, our objective was to determine whether and when analysis of published and unpublished placebo-controlled trials could have revealed cardiovascular risk associated with rofecoxib before its withdrawal, using the manufacturer’s trial data made available through the rofecoxib litigation, as an example to inform future post-market pharmaceutical safety surveillance efforts.
We included in our analyses all randomized clinical trials that compared rofecoxib with placebo and were completed as of termination of the APPROVe trial. Consistent with the manufacturer’s protocol for earlier meta-analyses of rofecoxib, we only included trials that examined adults using rofecoxib at 12.5 mg or more daily for 4 weeks or longer and excluded data from trial arms using rofecoxib in other doses or for shorter time periods.9–11 In addition, we only included data from subjects enrolled in the clinical trials who took at least 1 dose of the study drug (rofecoxib or placebo). We excluded data from 6 trials that did not include a placebo arm (Online Appendix, Figure A). In addition, we excluded data from 2 large ongoing trials that were terminated shortly after the APPROVe trial and the withdrawal of the drug: the Vioxx in Colorectal Cancer Therapy: Definition of Optimal Regime (VICTOR) trial and the Vioxx in Prostate Cancer Prevention (ViP) trial. At the time of termination, both of these trials had enrolled 25–33% of the planned number of subjects for a median treatment duration of 7.4 and 4.1 months respectively.12, 13 However, data from these trials were not pooled with APPROVe data at its termination to evaluate rofecoxib’s safety. A complete list of the randomized clinical trials included in this analysis is in Table 1.
In order to confirm that the data made available by the manufacturer were accurate, we compared the following characteristics of the data made available through the rofecoxib litigation with the data published within the trials’ manuscripts or data reported to the Food and Drug Administration (FDA) as part of each trial’s Clinical Study Report: number of study subjects, mean age, and proportion of women in each trial arm; this comparison is available as an Online Appendix, Table A1.
Our pre-specified outcome measure for this study was incidence of any investigator-reported death or cardiovascular thromboembolic (CVT) adverse event that occurred in an enrolled clinical trial subject. As pre-specified by the company, all trials collected adverse event data on subjects while they were either using the study medication (rofecoxib or placebo) or within 14 days after study drug discontinuation. In addition, studies 078, 091, and 122 collected adverse event data on subjects beyond 14 days after study drug discontinuation until study termination. For these trials, all available data were used in an intention-to-treat (ITT) fashion, the gold standard approach for analysis of clinical trial data.14
The randomized controlled trials conducted by the company used either Clinical and Regulatory Information Strategic Program (CRISP) or Medical Dictionary for Regulatory Activities (MedDRA®) terms for the reporting of trial adverse events, with approximately 50% of trials using each set of terms. CRISP is the company’s clinical data system medical terminology, and MedDRA® is the international medical terminology developed under the auspices of the International Conference on Harmonization of Technical Requirements for Registration of Pharmaceuticals for Human Use. We identified CVT events from lists of terms used by both the CRISP and MedDRA® systems. Two investigators (JSR and HMK) independently reviewed all adverse event terms used by the medical terminology systems and categorized those terms that represented CVT events without knowledge of the effect on the results. Disagreements were resolved by consensus. Examples of selected terms include acute myocardial infarction, embolic stroke, and unstable angina. The complete list of included CVT event terms is available as an Online Appendix, Table A2.
All statistical analyses used subject-level data and did not utilize pooling of summary information. We first calculated event rates and relative risks using Cox proportional hazards models stratified by study with treatment as the sole covariate. To replicate the manufacturer’s analytical practice,3 if there were fewer than 11 events in either study arm, the rate ratio was computed with the use of exact methods. We then cumulatively calculated event rates and relative risks using Cox proportional hazards models (or exact methods), pooling new data as clinical trials were completed and would be available for analysis. To account for the time necessary in practice to prepare and organize a final clinical trial data set for analysis, with 3 exceptions, data were only eligible for pooling 3 months after the ‘Last Person Out’ date of each randomized clinical trial (or the trial termination date; trial 126 was terminated early after trial 091 found rofecoxib to not be effective). Exceptions to this rule were made for trials 078 and 122. Because data from trial 078 were included in 2 pooled analyses of cardiovascular events by the company within Safety Update Reports required by the FDA while the study was ongoing, we likewise included data from trial 078 in our pooled analysis as of the same dates.15 The first Safety Update Report included all clinical trial data available as of March 16, 2001, and the second as of January 31, 2002. As of these Report dates, trial 078 was the only sizable placebo-controlled trial for which data were included while collection was ongoing. Similarly, data from trial 122, the APPROVe trial, were initially analyzed while the study was ongoing by the trial’s DSMB; these interim results were subsequently published,3 before completion of the long-term post-drug follow-up of participants.16 We likewise included data from trial 122 as of the date of the interim analysis, September 30, 2004.
For every fitted Cox model on complete datasets, we checked the proportionality assumption using a standard test based on the scaled Schoenfeld residuals. No failures of proportionality were noted. We did not include drug dose as a covariate in our analyses because more than 90% of the patient-years of observation were of patients using 25 mg of rofecoxib. All statistical tests were 2-tailed and analyses were performed using the R statistical software environment, Version 2.8.0 (http://www.r-project.org).
The work performed on this study was donated by the authors and there were no outside funding sources. All authors are currently or have been previously compensated for their participation in litigation related to rofecoxib at the request of plaintiffs and against Merck and Co., Inc. While the authors obtained access to the study data through participation in the litigation, no attorneys or others related to the litigation had any role in study conception or design, data analysis or interpretation, or the writing, review or approval of the manuscript, and the authors were responsible for the decision to submit the manuscript for publication.
We identified 30 randomized placebo-controlled trials with 17,256 subjects that met our inclusion criteria, 15 of which examined the efficacy of rofecoxib for treating osteoarthritis, 6 for treating rheumatoid arthritis, 3 for preventing or delaying the progression of Alzheimer’s disease, as well as 6 other trials, including the APPROVe trial (Table 1). To our knowledge, 24 of these 30 trials were published as manuscripts, although 6 (excluding APPROVe) were published after Merck withdrew rofecoxib from the worldwide market; therefore, data representing 36% of patients studied in placebo-controlled clinical trials prior to APPROVe had not been published prior to market withdrawal. Trial duration ranged from 4 weeks to 4 years, enrollment ranged from 17 to 2586 subjects assigned either rofecoxib or placebo, and rofecoxib dosage ranged from 12.5 mg to 50 mg, although more than 90% of the patient-years of observation were of patients using 25 mg of rofecoxib.
Through September 2004, there were 301 CVT adverse events reported by investigators during these 30 randomized placebo-controlled trials, 182 among 7034 patient years of rofecoxib use and 119 among 6695 patient years of placebo use (rate ratio [RR]=1.48, 95% confidence interval [CI], 1.17–1.87; p<0.001) (Table 2).
Through September 2004, there were 130 deaths from any cause reported by investigators during these 30 randomized placebo-controlled trials, 81 among 7158 patient years of rofecoxib use and 49 among 6805 patient years of placebo use (RR=1.71, 95% CI, 1.20–2.45; p=0.003) (Table 2).
Through September 2004, there were 372 subjects for whom an investigator reported either a CVT adverse event or death from any cause during these 30 randomized placebo-controlled trials, 221 among 6357 patient years of rofecoxib use and 151 among 5723 patient years of placebo use (RR=1.43, 95% CI, 1.16–1.76; p<0.001) (Table 2).
Cumulative pooled analysis of the 30 identified randomized placebo-controlled trials is displayed in Figure. As of December 2000, 21 (70%) of these trials had been completed, having enrolled 9884 subjects for 1749 patient years of observation to either rofecoxib or placebo, during which time 36 subjects experienced either a CVT adverse event or death from any cause. At this time, risk of either of these events was greater among subjects assigned to rofecoxib and the difference was borderline statistically significant (RR=2.18, 95% CI, 0.93–5.81; p=0.07) (Table 3). With the addition of data available through March 2001, 12,207 subjects had been observed for 4946 years. At this time, the risk of CVT adverse event or death associated with rofecoxib was sufficient to suggest a statistically significant association (RR=1.35, 95% CI, 1.00–1.96; p=0.05). The addition of subsequently collected data strengthened the statistical association. As of January 2002, 14,406 subjects had been observed for 7806 years and rofecoxib was associated with a 39% increased risk of CVT adverse event or death (RR=1.39, 95% CI, 1.07–1.80; p=0.02). Finally, as of the interim analysis of data from trial 122 (APPROVe), September 2004, 20,152 subjects had been observed for 17,310 years and rofecoxib was associated with a 43% increased risk of CVT adverse event or death (RR=1.43, 95% CI, 1.16–1.76; p<0.001).
Our cumulative pooled subject-level analysis of the data from all published and unpublished randomized placebo-controlled trials of rofecoxib demonstrates a progressing trend toward increased cardiovascular risk associated with the medication compared with placebo as early as December 2000, reaching a P value of 0.05 by June 2001, nearly 3 and a half years before the manufacturer voluntarily withdrew rofecoxib from the worldwide market. These findings are particularly compelling because as early as the late 1990s there were concerns about cardiovascular risk that emerged in the drug development process.1, 2 These analyses provide a roadmap for how drug safety can and should be assessed, particularly after a drug has been introduced into the market. Because the recently enacted FDAAA requires the public disclosure of trial results within the ClinicalTrials.gov database within 12–24 months of study completion, including both efficacy and safety outcomes, clinical trial data will be available to conduct comprehensive iterative meta-analyses independent of the FDA and manufacturers. Substantial amounts of clinical trial data that have rarely been fully utilized to understand drug efficacy or safety will now be available and can be used by independent investigators to complement and corroborate surveillance done by the FDA and the companies.
The data used in our study were not publicly available for analysis by independent investigators at the time rofecoxib was being marketed and have only now become available through litigation. One critical advantage of our approach was the use of subject-level data, which allows far more flexibility with respect to pre-specifying and defining the outcome of interest, using statistical methods to manage heterogeneity, and investigating specific subgroups of subjects as well as rare outcome events, a critical issue in pharmaceutical post-market safety surveillance. Currently, the data that will be available through the ClinicalTrials.gov database will be summary-level data, which may lack information necessary to investigate specific sub-groups or outcomes, particularly rare safety outcomes.
In 2006, the Institute of Medicine (IOM) released a report commissioned by the FDA to examine drug safety.17 Until recently, the centerpiece of the FDA’s pharmaceutical post-market surveillance has been the Adverse Event Reporting System (AERS), which combines voluntary reporting of adverse events by patients and clinicians and mandatory reporting of adverse events by manufacturers. The FDA traditionally relied on AERS because clinical trials conducted by industry as part of the new drug approval or indication application are often not large or long enough to evaluate safety outcomes. In addition, many of the large Phase IV randomized trials requested by the FDA at the time of drug approval to further investigate potential drug safety issues are not conducted for a wide variety of reasons, including feasibility, recruitment barriers, and cost. The IOM report included twenty five recommendations which together suggested that the FDA needed to better prioritize ongoing, systematic efforts to monitor drug safety during the product’s entire market life,17 such as the approach we have modeled. Nevertheless, identifying those safety outcomes which should be monitored prospectively through an iterative and timely meta-analysis will be challenging and should be driven by biological plausibility, safety signals that arise during the drug development process and during pre-approval studies, and systematic monitoring of surveillance systems such as AERS and other health care data sources through the FDA’s Sentinel Initiative, including Medicare Part D claims, the Department of Veteran Affairs, the HMO Research Network, and others.
Our analysis builds upon the prior work of Jüni et al., published shortly after rofecoxib was withdrawn from the market, that found that rofecoxib was associated with more than twice the risk of myocardial infarction by the end of 2000.7 First, our analysis used all published and unpublished data at the patient-level, whereas Jüni’s was restricted to published data at the summary-level. Moreover, Jüni’s cumulative meta-analysis examined the risk of myocardial infarction, whereas ours examines the risk of cardiovascular thrombotic event or death, a more comprehensive assessment of drug safety. In addition, Jüni’s meta-analysis included all randomized clinical trials, including trials that used NSAIDs as a comparator. The largest trial with the largest effect size was the Vioxx Gastrointestinal Outcomes Research Study (VIGOR) trial,18 which compared rofecoxib to naproxen. In fact, Jüni’s stratified analysis estimated a tripling of the risk associated with rofecoxib when compared with naproxen and a non-significant increase in risk when compared with placebo, although a test of interaction by comparator was not significant. Our analysis extends and strengthens these findings, demonstrating an increased cardiovascular risk associated with rofecoxib when compared with placebo.
There are several reasons why our analysis differs from those conducted by the manufacturer and published before rofecoxib’s market withdrawal.9–11 First, each of these pooled analyses focused on the Antiplatelet Trialists’ Collaboration (APTC) endpoint, which accounts only for cardiovascular, hemorrhagic, and unknown deaths, including hemorrhagic gastrointestinal deaths, as well as nonfatal myocardial infarctions and strokes (ischemic and hemorrhagic). The APTC endpoint is a composite outcome that was developed for the purpose of testing the net efficacy and safety of antiplatelet therapeutics that may prevent ischemic events but simultaneously may cause hemorrhagic events, specifically capturing overall benefits and risks of these types of medications, but is not appropriate for evaluating rofecoxib’s CVT risk. In contrast, we accounted for all ischemic-related CVT adverse events, as well as deaths from any cause. Accounting for all deaths is important for studies of cardiovascular risk, particularly among clinical trials not designed to examine cardiovascular outcomes, as cardiovascular deaths are not always apparent as such. In addition, nearly three-quarters of deaths in this analysis occurred within the three Alzheimer’s trials. Studying all-cause death is particularly important among patients with Alzheimer’s disease or cognitive impairment who may not report symptoms accurately and may not always be as thoroughly evaluated when brought for medical attention. Second, we examined all available data from all 30 trials with a placebo-controlled arm, including data not used in company-conducted analyses9–11 (Online Appendix, Table A3), although we did exclude data from the VICTOR and ViP trials, two large ongoing trials that were terminated shortly after APPROVe and the decision to withdraw the drug. Finally, we accounted for all adverse events that were investigator-reported among all trials and included outcomes observed beyond the 14-day window after drug discontinuation in an ITT fashion, the gold standard approach for analysis of clinical trial data14 and the approach pre-specified within these trials’ data analysis plans.
Our analysis raises the issue of how investigators should interpret data relevant to drug safety. Although we want to avoid false positive concerns, conventional standards of statistical significance may be too strict for issues of safety.19, 20 The importance of a safety finding depends on many factors, including plausibility of harm, effect size, number of people exposed, and alternatives to treatment. Understanding the range in the estimation of the risk may be more important than focusing on hypothesis testing using a p value threshold for significance of 0.05.19 For instance, at the time of the publication of the VIGOR trial in November 2000,18 4 years before the drug was withdrawn from the market, when concern about rofecoxib’s cardiovascular risk heightened and there were clear alternatives for treatment of osteoarthritis and rheumatoid arthritis, our analyses suggest that rofecoxib was associated with more than a doubling of the cardiovascular risk, with 95% confidence intervals that ranged from 0.93 to 5.81 but a p value of 0.08. This indication of risk needed to have been interpreted alongside the 5-fold increased risk of myocardial infarction among rofecoxib users found in the VIGOR trial,21 as well as with the concerns about cardiovascular risk that emerged in the drug development process.1, 2
Physicians and the public deserve to be in a position to make informed choices about risk and benefits and the early disclosure and dissemination of information about potential risk after its recognition must be required. Our study provides insight into what should have been known about the risks of rofecoxib. The signal of cardiovascular risk appeared soon after the drug was FDA-approved and made available in May 1999. But more importantly, our study suggests that such analyses should be ongoing for all drugs for which trials are being conducted, with attention to the rapid addition of new data about potential harm to any cumulative pooled analyses. Ideally, subject-level data should be made available to the public to allow for independent assessments. If we are to detect harms early and protect the public’s health, while ensuring the availability of new, clinically-effective therapeutics, a system needs to be established that makes full use of all existing evidence.
The authors would like to thank Dr. John B. Kostis for his assistance in selecting cardiovascular thromboembolic event terms from the CRISP and MedDRA® event systems. Dr. Kostis received no financial support for this work. MedDRA® is a registered trademark of the International Federation of Pharmaceutical Manufacturers and Associations (IFPMA).
Financial Disclosures: Dr. Madigan is currently a consultant at the request of plaintiffs in litigation against Merck and Co., Inc. related to rofecoxib in Australia and was previously a consultant in litigation in the United States. All other authors were previously consultants at the request of plaintiffs in litigation against Merck and Co., Inc. related to rofecoxib in the United States. Over the past 5 years, Dr. Madigan has been a consultant to Pfizer, Wyeth, Sanofi-Aventis, and Takeda and currently serves on the clinical review team of iGuard.org. Dr. Krumholz has had research contracts with the American College of Cardiology and the Colorado Foundation for Medical Care; has previously served on the advisory boards of Alere and Amgen, and currently serves on one with UnitedHealthcare; is a scientific advisor for Centegen; has been a subject expert for VHA, Inc.; has received speakers’ compensation from the American College of Cardiology; and is Editor-in-Chief of Circulation: Cardiovascular Quality and Outcomes, and Journal Watch Cardiology of the Massachusetts Medical Society.
Funding/Support: This project was not directly supported by any external grants or funds. Dr. Ross is currently supported by the National Institute on Aging (K08AG032886) and the American Federation of Aging Research through the Paul B. Beeson Career Development Award Program.
Author Contributions: Drs. Ross, Madigan, Wang, and Krumholz had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the analysis.Study concept and design: Ross, Madigan, Egilman, Krumholz
Acquisition of data: Madigan, Wang, Krumholz
Analysis and interpretation of data: Ross, Madigan, Hill, Egilman, Wang, Krumholz
Drafting of the manuscript: Ross, Krumholz
Critical revision of the manuscript for important intellectual content: Ross, Madigan, Hill, Egilman, Wang, Krumholz
Statistical analysis: Madigan, Wang
Administrative, technical, or material support: Krumholz
Study supervision: Krumholz