How to Feed a Lawyer (and Other Irreverent Observations from the Legal Underground)

    Click on the book cover for details!

« | Main | »



I know Evan's just baiting me, but forgive me if I'm not persuaded that plaintiffs have shown Merck's theory is a myth (or even that their expert testimony will satisfy Daubert once the Fifth Circuit gets around to reviewing Fallon's opinion) when the judge's ruling admitting the testimony called it "wholly conclusive, rather than explanatory." Is there any appellate court that allows an expert to make a "quantum leap" from data to conclusions and survive Daubert? I know Fallon's characterization alone is enough to keep evidence out in the Seventh, e.g., Rosen v. Ciba-Geigy (7th Cir. 1996), and I didn't see him cite any Fifth Circuit precedent in support of admission.

As long as we're talking Vioxx, I wonder if Evan can help me solve the mystery of why Mark Lanier hasn't moved to enter judgment in Ernst?


In fact, why not call it what it is: junk science.

Ah, so since Fallon also admitted Merck's expert opinions in favor of the eighteen-month theory, Evan is effectively conceding that Fallon's decision to admit evidence into trial doesn't tell us whether the theory is any good. So why trumpet the admission of Dr. Gandy's testimony?


Ted: "Junk science"--I'm just trying to take that term back from the tort reformers.

As for Judge Fallon's ruling, I'm aware that there are other ways to slice it. Anyone interested should read the opinion, which I linked to.

Finally, as to the Lanier mystery, I don't know. I saw the post at Point of Law about it but didn't have an answer.


Thanks ATLA! Keep fighting the good fight for the little guy.

King of the Cows

And we all know that trial judges are infallible, right?


King: Federal trial judges are infallible. President Bush taught me that.



I think it's -- well, let's call it a little misleading -- to equate Grandy's testimony with the plaintiff's expert testimony. True, Grandy is a clinical doctor, not a researcher, but the plaintiff has at least two eminent researchers who will testify about general causation: professors at top universities with hundreds of publications.

I see from your blog that you don't understand this distinction between clinicians and researchers. You criticize University of Michigan Professor Lucchesi for "sloppiness" because he "never looked at the slides of Irvin's blocked artery." But Lucchesi isn't a clinical doctor, he's a professor of pharmacology. So of course he's not going to testify out of his area of expertise. Ted, you owe this eminent researcher an apology for your ignorant slur.


But Lucchesi isn't a clinical doctor, he's a professor of pharmacology. So of course he's not going to testify out of his area of expertise.

But he did testify out of his area of expertise by saying that Vioxx was highly likely to have caused Irvin's heart attack. The statistical evidence doesn't support that: there are no studies showing a high relative risk for low-dose Vioxx taken over short periods of time, and even the studies of long-term low-dose Vioxx use don't show a high enough relative risk factor to justify the characterization "highly likely." So Lucchesi can only reach that result by ruling out other risk factors—which, as you have just argued, he's not qualified to do. He's jumped to a conclusion, and Daubert isn't supposed to allow that.



I read your post on, where you try to debunk Lucchesi's medical testimony. There's little point in trying to rebut you, since University of Michigan Professor Lucchesi is a world-renowned expert, and you're a lawyer. Similarly, three judges have now admitted Lucchesi's testimony, despite their legal obligation to exclude junk science from the courtroom. In my view the medical opinions of top experts like Lucchesi and Stanford professor of medicine Farqahur (scheduled to testify for the plaintiff) are a lot more credible than yours.

Still, neither of these top scientists are here, so let me address one sentence of your post, which contains two major errors -- errors you repeat throughout your post, and throughout your writing on Vioxx.

You say: "As you recall, the VIGOR study showed a 1.92 relative risk of heart attack for those who had taken Vioxx for eighteen months."

Your first error is that the VIGOR study actually found a relative risk of 4.5 for heart attack, not 1.92. That is, Vioxx more than quadrupled the risk of a heart attack. The 1.92 number you cite is for heart attack plus stroke plus other bad stuff. This is a key point since, in many studies, Vioxx seems to increase the risk of heart attack more than the risk of strokes and such. But Merck has usually combined all these various outcomes in their published studies, lowering the overall risk they report -- and they've been criticized for this practice in leading medical journals. This is also a key error because most people suing Merck have had heart attacks, not strokes or other awful stuff.

The second error in the sentence I quote is your claim that the VIGOR trial only found Vioxx dangerous after 18 months. In fact, the VIGOR study found that Vioxx started raising risks after two months. It said nothing at all about risks after 18 months, because the study only lasted 12 months.

Let me repeat that: the VIGOR trial did not find Vioxx safe after 18 months. Contrary to your claim, it did not find Vioxx dangerous after 18 months. The VIGOR trial ended after 12 months.

Here's the FDA's summary of the findings from Merck's VIGOR trial:

"...the Vioxx GI Outcomes Research (VIGOR) trial in which rofecoxib 50 mg once daily was compared to naproxen for up to 12 months. In VIGOR, rofecoxib was associated with a hazard ratio of approximately two compared to naproxen based on the composite endpoint of death, MI, or stroke. In contrast to the findings in APPROVe, in VIGOR the Kaplan-Meier CV event curves for rofecoxib and naproxen began to separate after approximately two months of treatment." [from page 5 of the FDA Memo I linked to]

So Ted, let me suggest that you refrain from opining about medicine and science, and stick to topics you actually know something about, like the law. And let me repeat that you owe Lucchesi an apology.


Ragout is correct that I should have said APPROVe, rather than VIGOR (which was a high-dose study), which resolves most of the criticisms he has of my post. I've corrected that error. I do appreciate his honest link to the FDA memo, which the plaintiffs have repeatedly tried to exclude from jury consideration. Ragout is wrong or misleading in other material respects, and I stand by the edited post.



If your only error were confusing VIGOR and APPROVe (probably the two most important Vioxx studies), I might have chuckled, but I wouldn't have written such a long post. Even with your corrrections, your two errors remain.

First, you report findings for "cardiovascular events," and imply that this means "heart attacks," which is not true. APPROVe, like VIGOR and many other studies, found that Vioxx raises the incidence of heart attacks by much more than stroke and other "cardiovascular events." See this Lancet study for some cogent criticism of Merck's practice of conflating various cardiovascular illnesses rather than focussing on heart attack. Recall that Merck is mostly being sued by people with heart attacks, not strokes.

Second, you're now ignoring the VIGOR trial's findings (elevated risk starting in month 2) rather than misreporting them. This is a slight improvement, but it would be much more intellectually honest if you were to acknowledge that VIGOR (and many other studies) found that Vioxx elevates risk after very short exposure.


VIGOR involved high-dose rofecoxib. Irvin was taking only 25 mg.

High-dose rofecoxib appears to significantly increase risk of cardiovascular events after sixty days of exposure. My understanding is that Merck advised that high-dose usage should go no longer than thirty days. The FDA memo you cite (and that plaintiffs try to hide from juries) says "Short-term use of NSAIDs to relieve acute pain, particularly at low doses, does not appear to confer an increased risk of serious adverse CV events (with the exception of valdecoxib [Bextra] in hospitalized patients immediately post-operative from coronary artery bypass (CABG) surgery)."

Ragout protests that I use a too general level of specificity through "cardiovascular events." He prefers "heart attacks." But if one goes even more specific, to "fatal heart attacks," we're back to a relative risk that is indistinguishable from 1, even after eighteen months of use. Why cherry-pick? What's the mechanism that increases non-fatal heart attacks, but not other cardiovascular events including fatal heart attacks, and from that, why can we conclude that that the number implicates Irvin's use of Vioxx?


Side note: Ragout suggests that I'm not capable of evaluating Lucchesi's testimony because I'm just an attorney, and therefore my commentary is beneath rebuttal. (I don't know if his opinion would change just because I'm an attorney with extensive and graduate-level statistical training, but let's assume that it wouldn't.) Perhaps this is so. But, then, what does it say of our legal system that the people who will be charged with evaluating Lucchesi's testimony will be even less-qualified to do so than I am?



It's not just my opinion that Vioxx causes heart attacks but has less effect on strokes: it's the opinion of Juni et al, the authors of the meta-analysis in the top medical journal that I linked to. As the article makes clear, they base their opinion not on "cherry-picking" a few findings, but on a thorough review of the literature and an understanding of the medical science. They discuss the mechanisms through which Vioxx seems to cause harm, and point to research on other drugs that act on the same mechanisms. This research found that these drugs had differing effects on heart attacks and strokes. Hence, they conclude that it was "misleading" of Merck to group these (and other) different outcomes together.

I could go on, but your confusion about the FDA memo is much more interesting. If you actually read the memo, instead of just the executive summary, you'd see that the FDA is not ruling out very small risks (specifically, risks too small to be detected by clinical trials). They say that the "CV risk of short-term use of NSAIDs is very small, if any." (Note that by short-term use, the FDA means one or two weeks, so I don't see how this supports your belief that Vioxx is safe for 18 months).

Anyway, the FDA's statement is perfectly consistent with Lucchesi's expert testimony that a single pill can cause a heart attack. It's quite possible that Vioxx can triple the risk of a heart attack, but still be safe to take for a week or two during an episode of particularly acute pain.

To see why, consider some numbers. Suppose that someone has 2 chances in 100,000 of having a heart attack on a given day. This is a perfectly plausible number for a 50-60 year old man -- it equals about 3/4 of a percent per year. Further suppose that each Vioxx pill triples the risk of a heart attack for a day, also a perfectly plausible figure. That is, on days when you pop a Vioxx, your chance of a heart attack is 6 per 100,000. Given these assumptions, it might be perfectly reasonable to take Vioxx for a week or two, raising your chances of a heart attack by 4 in 100,000 per day. No clinical trial is every going to be big enough to reliably detect this risk over a few weeks. But over longer periods of time, the danger really adds up: over a year, taking Vioxx would raise your risk of heart attack from under 3/4 of a percent to over 2 percent.

This shouldn't be surprising. I've simply made the mundane point that even tripling a risk isn't too bad, if you start from a small enough base. To take one example, parachuting is safe enough that people do it for fun. But try parachuting every day for a year, and I bet the odds are pretty high that one day your chute will fail to open. Apparently the FDA understands this point even if their memo has failed to convey it to a highly-educated person such as yourself. (Perhaps the memo will confuse the jury too?)

But if each Vioxx pill triples the risk over the course of a day, then 2/3 of the people who die from a heart attack while taking Vioxx -- even after one day -- have died from a heart attack caused by Vioxx. These are heart attacks for which Merck should be liable.


But if each Vioxx pill triples the risk over the course of a day, then 2/3 of the people who die from a heart attack while taking Vioxx -- even after one day -- have died from a heart attack caused by Vioxx.

But if each Vioxx pill triples the risk over the course of a day, we'd see statistical evidence of it. And we don't. We barely see any increase in risk for 25 mg Vioxx. Even David Graham's gigantic retrospective study found a 25% increase, which bordered on statistical insignificance, but even giving the 1.25 relative risk the benefit of the doubt means that 4/5 of the people who have a heart attack while taking Vioxx would have had the heart attack anyway; he had to juice his numbers to get to 140,000, and I lost all respect for The Lancet for publishing the 140,000 number.

No clinical trial is every going to be big enough to reliably detect this risk over a few weeks.

Exactly. Which is why holding Merck liable with 20/20 hindsight is insanity.


"But if each Vioxx pill triples the risk over the course of a day, we'd see statistical evidence of it. And we don't." This is where you are wrong.

In the kind of trials Merck ran (randomized trials on a few thousand people screened to have a low risk of a heart attack), it is very unlikely that you'd find evidence of elevated risk in a short period of time. One can formalize this and say that the APPROVe trial lacks statistical "power" to detect a tripling of the risk of heart attack even after two years. I've done the calculations: they're not hard, but I bet that the notion of statisical "power" didn't come up in your "graduate-level" statistical training. "Power" is a crucial statistical notion in epidemiological studies.

This is what I was trying to explain in my long post above. In the APPROVe trial, the controls had 12 heart attacks over 3 years. Assuming that Vioxx triples the risk, we'd expect the treatment group to have 36 heart attacks (actually they had 31). If this risk kicks in from day one, the most likely outcome after 3 months (never mind 3 days) is 1 heart attack for the controls and 3 for the Vioxx group. Needless to say, this would not be statistically significant with the APPROVe sample sizes. You'd get a 95% confidence interval of (0.24,157)! That is, even though I've made up the data under the assumption that the true relative risk is 3, and assumed that the most likely outcome is what actually happens, you'd still get a super-wide confidence interval.

For these reasons, we only see statistical evidence of Vioxx's harmfulness in the first few weeks in some very large observational studies, and studies where the risk of heart attack was already elevated (e.g., studies of post-surgery patients).

On your other point, it just boggles my mind that you would criticize Graham for relying on randomized trials in calculating 140,000 excess deaths, rather than his own study, which is a less reliable "observational study." Similarly I'm puzzled that you criticize Prof. Ray for testifying not only about his own observational study, but instead relying on all the Vioxx studies. Perhaps you could sketch out your reasoning? Do you think that scientific studies are like legal briefs where you make the strongest possible arguments to support a pre-determined conclusion? Unless you believe something crazy like that, I don't see how you can possibly have any problem with Graham and Ray.


You make three points, but each of them address straw men, rather than my actual argument.

1. I understand the issue of statistical power. I question your relative risk figure of 3, and the implicit hypothesis that APPROVe (not to mention every other study) isn't powerful enough to find a three-fold increase eighteen-months in. The fact of the matter is that controls had identical rates of CV events up through eighteen months--at the nine-month stage, controls had more CV events. That's not inconsistent with a risk rate of 1.25 or 1.5 (though it's also not inconsistent with a risk rate of 1, and the plaintiff has the burden of proof to reject that null hypothesis), but it is inconsistent with a risk rate of 3, which no study has shown for low-dose Vioxx.

2. it just boggles my mind that you would criticize Graham for relying on randomized trials in calculating 140,000 excess deaths

I didn't do that. I criticized Graham for cherry-picking two different randomized trials (one for low-dose and one for high-dose) that gave him the highest number for his calculation. In his internal FDA memo, he used a lower excess-death number. By the way, Graham's number is "excess heart attacks", not "excess deaths." For whatever reason, there's still no evidence of a higher risk rate for cardiac deaths for Vioxx use. My suspicion (subject to revision in the face of additional evidence) is that it's a statistical quirk: the real risk rate is perhaps somewhere between the CV rate of 1.92 and the statistically insignificant CV-death risk rate of 0.9 (0.3 for fatal MIs in APPROVe).

3. Ray didn't "rely on all of the Vioxx studies." For example, Ray omitted the Alzheimer's study showing no elevated CV events. Had he included this study, the same methodology would have shown that Merck should have done something after APPROVe, rather than after VIGOR: i.e., in 2004. The point remains that his claim on the stand is that Merck "should have known in 2000," but he had access to the same information that he said Merck should have relied upon, but published a paper with a different prospective conclusion in 2001. That's a litigation opinion, not a scientific opinion.


1. I'm not hypothesizing about the low power of the APPROVe trial, I've done the calculations. The APPROVe trial found a relative risk for heart attack of 2.8 for 25mg Vioxx. The power of the APPROVe study is 74% after 2 years, against the alternative hypothesis of a 2.8 RR. This counts as "underpowered," since the usual rule of thumb is to aim for a power of 80% or higher. In other words, even assuming an RR of 2.8, there is a 26% chance that if the APPROVe trial were run again for 2 years, no statistically significant risk would be found. Under your hypothesis that Vioxx doubles the risk, the power would be much lower, of course.

2. My guess is that Graham's number changed between the draft and the published article because an editor asked Graham to base his calculations on randomized trials rather than his own observational study. This is called "peer-review."

You're right that the Graham number is for excess heart attacks, not deaths. The distinction between heart attacks and deaths from heart attacks nicely illustrates why your method of interpreting these studies is naive. Unless Vioxx causes some special kind of mild heart attack, the RR of heart attacks and deaths from heart attacks should be the same. The medical researchers don't seem to know of any evidence that Vioxx heart attacks are any different from the usual kind, so my assumption is that the RRs are the same. In addition, I believe that it has been well established that one heart attack puts you at greater risk for a second (which may be fatal). The Vioxx trials did't count second heart attacks even when they occured within the study period.

So you need to interpret the results of a clinical trial in the light of other medical knowledge. You speak as if the APPROVe trial were the only evidence there is.

In interpreting the studies, one should rely on heart attacks rather than deaths from heart attacks, because heart attacks are more common. Hence statistical tests will have greater power to detect heart attacks than deaths from heart attack.

3. The Ray paper you linked to did not make any overall conclusion about the safety of low-dose Vioxx. It wasn't a meta-analysis (of many studies). It was merely a report on the results from on particular study.

I'm not sure on what basis you conclude that Ray "omitted the Alzheimer's study." Newspaper reports?


Ray in 2001: "there was no evidence of raised risk of CHD among users of rofecoxib at doses of 25 mg or less or among users of other NSAIDs."

This was linked in my original post, so we're repeating ourselves. You can have the last word.


I'll take the last word.

In the line you quoted, Ray was talking about the results of one particular study. He wasn't drawing general conclusions about Vioxx.

The comments to this entry are closed.

My Photo

About Evan Schaeffer

Search Beyond the Underground