Starlight Temple

The New England Journal of Medicine has published a direly-worded Expression of Concern about a Merck paper on Vioxx's heart risks. They claim Merck withheld data from the paper, and insinuate that secret evidence shows Merck deleted it at the last moment before publication. They finish by saying "Taken together, these inaccuracies and deletions call into question the integrity of the data on cardiovascular events in this article. We have asked the authors to submit a correction to the Journal."

Sounds bad for evil ol' Merck, eh? Well have a look at the NEJM's own data:

There are some major problems here.

Problem 1: Confidence ranges as wide as the Grand Canyon. This is not surprising, since the study produced less than two dozen measurable events. The results are just too small to draw precise conclusions. About all you can say with any confidence is that Vioxx looks fairly scary.

Problem 2: Alleged missing data makes no clinical difference. No doctor said "Oh no! If I had known the relative risk as 20 instead of 17, I never would have prescribed the drug!" It makes no practical difference. The data show a clearly risky drug, with a few percent chance of being spectacularly dangerous.

Problem 3: Secret evidence. The NEJM claims "We determined from a computer diskette that some of these [relevant adverse cardiovascular] data were deleted from the VIGOR manuscript two days before it was initially submitted to the Journal on May 18, 2000." What diskette? Of what origin and chain of custody? Determined how? By whom? In accordance with which published forensic data analysis standards? The NEJM is not saying.

This is not science. It isn't even journalism.

Problem 4: Uncontrolled data. At least from what the NEJM is showing here, the original trial had no placebo arm. It only compared one drug with another, not with no drug treatment at all. Merck told the FDA that naproxen must have been protecting people against heart attacks, and that the more frequent rofecoxib events were just the underlying heart disease of the study group. Without a placebo subgroup, who's to know? Skeptical scientists, that's who. Crap like this would get a bad grade in undergraduate biology. The FDA and the original NEJM reviewers should have sent Merck back for a do-over.

Lies. Damn lies. Statistics. Medical trials without a placebo subgroup.

Problem 5: Activism, not accuracy. Speaking to the New York Times, one of the NEJM article's authors says

"They did not disclose all they knew," Dr. Curfman said. "There were serious negative consequences for the public health as a result of that."

Say what?! The original data showed Vioxx to probably be dangerous, and the alleged data makes no material difference in that conclusion. Furthermore, Curfman provides no evidence that the data were hidden from the FDA, or witheld from the Prescribing Information sheet.

The NEJM appears to be outpacing The Lancet in the political activism olympics. I hear the winner gets a gold-plated Sphincter Medallion.

UPDATE: One of Derek Lowe's commentors, which I somehow missed earlier, makes many of the same points:

So are calculations made "incorrect"? Only in a way that an innumerate reader would care about. ... I doubt this "Editorial" would have made it through peer review without changes. I wonder what process NEJM used to assess it before going ahead and publishing? And will they be reconsidering that process going forward?