The New England Journal of Medicine published an analysis to the APPROVe data that concludes "one could not conclude from the data that a shorter course of rofecoxib is safe."
Merck disagrees. I take no position at this time, because I don't know whether Merck is right that linear time interaction (which shows no significant increased risk for Vioxx before 18 months) makes more scientific sense than the logarithmic time interaction reported by NEJM that provided results that barely missed the .05 significance range. (I am inclined to agree with Jeffrey Drazen's statement that one sticks with the method specified in the paper in advance—though NEJM had no problem raking Merck over the coals earlier for adhering to pre-approved scope of study rather than updating data (Dec. 8; Feb. 22; Feb. 27). It's also entirely possible that NEJM is right to make this "correction" (on grounds of sticking with pre-approved criteria) and Merck is correct that the best scientific view of the APPROVe data is that short-term usage does not have a higher relative risk (on grounds that the pre-approved criteria do not provide the best explanation for the data).)
It's a complex statistical issue, so I'm surprised to see Evan Schaeffer state with confidence that this proves Merck wrong. I look forward to Schaeffer's scientific analysis of why linear time is the inappropriate means of evaluating the data, something he inexplicably omitted from his post defining Merck's scientific argument as a myth. And as Derek Lowe notes, even the logarithmic statistical evidence is ambiguous at best—which, in a court system where the rules for evidence were being followed, would count against the party with the burden of proof.
What I am certain of is that other pro-plaintiff websites are misrepresenting the data. The Settlement Channel claims "A few months of Vioxx can likely cause cardiac harm." But NEJM says no such thing: it certainly does not say that a few months of Vioxx is "likely" to cause cardiac harm (the odds of such harm remain remote); nor does it say that one suffering a cardiac problem is more likely than not to have had that harm caused by Vioxx (the relative risk is well below the 2.0 required to make that "likely" judgment for short-term low-dosage use). All NEJM said was that one cannot conclude with statistical certainty, using one particular statistical test, that short-term Vioxx use is safe. That's a far cry from "likely" on either general or specific causation for any given plaintiff.
It's a sign of how far from science the tort system is that plaintiffs' attorneys with the burden of proof are treating such a statement of ambiguity as affirmative, rather than negative, evidence. (Then again, such bravado may just be another example of the plaintiffs' bar's incentive to spin, which I've earlier discussed.)