No Pain Relief for Tort Sufferers

James Hamilton takes a look at one of the key studies on Vioxx and heart attacks.  He is not greatly impressed.

I took a look at one of the studies on which the decision was justified, written by Dr. David Graham and co-authors and published in Lancet in February. This study looked at 8,143 Kaiser Permanente patients who had suffered a heart attack and had also at some point taken a nonsteroidal anti-inflammatory drug (NSAID), of which Vioxx (rofecoxib) is one. Of these patients, 68 were taking rofecoxib while 4,658 were receiving no medication at the time of their heart attack, a ratio of (68/4658) = 1.46%. For comparison, the study looked at 31,496 other patients who had also at some point taken an NSAID, matched for characteristics like age and gender with the first group, but who didn't have a heart attack. The ratio of rofecoxib users to those with no current medication was slightly lower (1.05%) in this second group, which one might summarize as a (1.46/1.05) = 1.39-fold increase risk of heart attack from taking rofecoxib compared to no NSAID. Is that statistically significant, in other words, can you rule out that you'd see a difference of that size just by chance? Yes, the study claimed, but just barely.

On the other hand, this was not a controlled experiment, in which you give the rofecoxib randomly to some patients and not others in order to see what happens. Rather, something about either these patients or their doctors led some of them to be using rofecoxib and others not. Dr. Graham and co-authors looked at a variety of indicators that suggested that the rofecoxib patients already had slightly elevated risk factors for coronary heart disease. Once they controlled for these with a logistic regression, their study found an elevated risk factor of heart attack for rofecoxib takers of 1.34, which was not statistically significantly distinguishable from 1.0.

The strongest evidence from this study was a claimed dose-effect relation. Of these 68 rofecoxib-using heart-attack patients, 10 of them were taking doses above 25 mg per day. Only 8 patients in the much larger control group were taking so high a dose, implying an elevated risk factor of 5 to 1 for high-dose patients. Again observable risk factors could explain some of this, with the conditional logistic regression analysis bringing the implied drug-induced risk down to 3 to 1. According to the study, this elevated risk factor was still statistically significant, even though the inference is based on the experience of just 10 patients.

The obvious question here is whether in fact the authors were able to observe all the relevant risk factors. The study openly acknowledged that it did not, missing such important information as smoking and family history of myocardial infarction.

...[E]ven if there actually is an elevated risk of the magnitude the studies suggest but can't prove, the question is whether I might want to accept a 1 in 4,000 risk of dying from a heart attack in order to get the only medication timt makes my pain bearable and a mobile life livable.  And if I say no to the Vioxx, I may end up taking something that is less effective for my pain but has risks of its own.

.... How did we arrive at a system in which 12 random Texans are assigned responsibility for evaluating the scientific merits of statistical evidence of this type, weighing the costs and benefits, and potentially sending a productive blue-chip American company into bankruptcy protection?

See also my op-ed Bringing the Consumer Revolution to the FDA.