Boston, MA - Editors of the New England Journal of Medicine say the trial that prompted Merck to pull its popular COX-2 inhibitor rofecoxib (Vioxx) off the market had important errors. Prompted by a public announcement by Merck about potential statistical anomalies in the work, the editors investigated the data and issued a correction that is slated to appear in the July 13, 2006 edition of the journal [1,2]. In the reported results, the test for proportionality of hazards used linear time rather than the logarithm of time that was specified in the methods section.

Dr Curt Furberg

"They overinterpreted survival curves to conclude there was an 18-month delay in cardiovascular risk," Dr Curt Furberg (Wake Forest University School of Medicine, Winston-Salem, NC) told heartwire. "In their test for proportionality, they used the wrong p value and the data do not include an intention-to-treat analysis. But that is being corrected now," Furberg, who has served on Food and Drug Administration committees evaluating coxibs, said. "The error is troubling, and I don't know why it happened."

"The Vioxx cardiovascular data analysis plan called for numerous statistical and graphical methods to be used to assess whether the relative risk of Vioxx compared with placebo was constant over time or if it changed over time," Merck told reporters. "The reference to logarithm of time in the description of methods published in the New England Journal of Medicine and submitted to regulatory agencies was in error."

They overinterpreted survival curves to conclude there was an 18-month delay in cardiovascular risk.

The company says the reported result, with a p value of 0.01, came from a statistical model using linear time, not logarithm of time. "Recent tests show that the result using logarithm of time has a p value of 0.07," Merck reports. "Therefore," journal editors point out in their correction, "statements regarding an increase in risk after 18 months should be removed from the abstract."

This is not the first time the journal has issued a correction on rofecoxib data. In December, editors recommended changes to cardiovascular data in the largest rofecoxib study—the Vioxx Gastrointestinal Outcomes Research (VIGOR) trial. Now problems appear in the publication of the Adenomatous Polyp Prevention on Vioxx (APPROVE) trial.

"This latest correction is not something that could have been picked up by a review panel," executive editor Dr Gregory Curfman told heartwire. "Our editing procedures are rigorous, tight, and vigilant." He says that short of recalculating results, many details have to be left to the integrity of study authors. Curfman and managing editor Dr Stephen Morrissey did not speculate as to whether or not they were misled, but they note that the researchers report having made an honest mistake.

"At present, these investigators and the company don't look very good in the eyes of the medical community," Furberg said. "This illustrates the difficulty for reviewers. It demonstrates how authors have total control over what is reported, and reviewers and journals are at a disadvantage."

"I'm losing some trust in industry-sponsored trials, and I think we have to be very careful," Furberg said. "There are more data to come, and we will also be looking at drugs such as celecoxib [Celebrex, Pfizer]; so far, the data there are suggesting it is not very different."

In a letter to the editor, Dr Steven Nissen (Cleveland Clinic Foundation, OH) points out the Adenoma Prevention with Celecoxib (APC) trial, similar in design to APPROVE, reported data in which the intention-to-treat approach was used with no censoring of delayed events [3].

"Since patients who stopped the study drug early are likely to be people who had adverse reactions, such as hypertension, heart failure, or renal dysfunction, they represent a particularly vulnerable group. It is now clear that the approach of censoring events that occurred more than 14 days after drug discontinuation had a significant effect on the results of the APPROVE trial," Nissen writes. "In a report of a serious drug-safety problem, even if the original study design prespecified censoring of late events, it is particularly important to provide alternative analyses if such analyses suggest a substantially different conclusion."

Lead author Dr Robert Bresalier (University of Texas MD Anderson Cancer Center in Houston) was not available for comment, but in a response released by the journal, he and coauthor Dr John Baron (Dartmouth Medical School in Hanover, NH) reply to letters written by Nissen and Furberg [4,5].

"Dr Nissen characterizes the censoring of event data after 14 days as 'unusual.' Our impression is that such follow-up is actually common and is usually conservative since it avoids the dilution of a toxicity signal that may occur when an active drug is discontinued," they write. "Dr Furberg implies that our analysis was not an intention-to-treat analysis. This is true, in the sense that we did not follow patients more than 14 days after they discontinued treatment. However, all cardiovascular events observed during the study follow-up were assigned to treatment groups."

The scientific debate surrounding the APPROVE study in no way changes the company's commitment to defend the Vioxx litigation on a case-by-case basis.

Kenneth Frazier, senior vice president and general counsel of Merck, told reporters that the scientific debate surrounding the APPROVE study in no way changes the company's commitment to defend the Vioxx litigation on a case-by-case basis. "The individual facts of each case are very important. Unfortunately, thrombotic cardiovascular events are a major health problem and stem from many risk factors unrelated to Vioxx."

In a statement to the press, Peter Kim, president of Merck Research Laboratories, said, "As we have said previously and reaffirm today, this correction did not change the data in the APPROVE study or its results. It is important to understand that the correction centered on the description of a single statistical method."

Kim says the company used a number of analytical and graphical methods to evaluate whether the relative risk changed over time. "Taken together and consistent with good statistical practice, these methods confirm that the relative risk changes over time. In the APPROVE base study, the increased relative risk was observed beginning after 18 months."

But Curfman disagrees. "We never believed that in the first place, which is why we never allowed that statement to be made." Curfman says that many have misinterpreted the cardiovascular data to suggest that risk changes after 18 months. "This correction casts even further doubt on that conclusion."