- The American Conservative - https://www.theamericanconservative.com -

“When Half a Million Americans Died and Nobody Noticed”

Such was the provocative title under which Alexander Cockburn ran a recent column [1] discussing my China/America article [2] in The Week, a British-based news magazine which claims a total American print circulation of over 500,000. We’ll see whether anyone notices that column either.

Cockburn’s question referred to my examination of the American mortality figures [3]surrounding the heavily-promoted anti-pain drug Vioxx, released by Merck in 1999 and pulled from the market in 2004 after a published FDA study indicated it seemed to double the risk of heart attacks and strokes and had probably been responsible for at least tens of thousands of American deaths. I had noted that the major shifts in total American mortality bracketed by Vioxx’s introduction and recall—shifts which were concentrated in exactly those age-groups taking Vioxx and were due to the aforementioned heart attacks and strokes—may actually point to a total death-toll an order-of-magnitude greater than that initial scientific estimate. Continue reading → [4]


This column of mine column appears on my new personal website [5], which collects hundreds of my writings from the last couple of decades across a wide range of topics and categories, including:

Comments Disabled (Open | Close)

Comments Disabled To "“When Half a Million Americans Died and Nobody Noticed”"

#1 Comment By peter_schaeffer On May 14, 2012 @ 11:10 am

A few more notes
1. Vioxx
volumes declined by 30% from 2003 to 2004. For the entire year of 2003, Vioxx
was dispensed at a rate of 1.663 million prescriptions per month. For the first
9 months of 2004, the rate was 1.555 million prescriptions per month. That’s a
decline of 6.5%. Of course, after 2004-09-30 the prescription rate was zero.
As the
reader can see the decline in Vioxx usage, pre-recall was quite small.
Note that Vioxx volumes peaked in 2001 and declined thereafter. Why is not
clear. However, Bextra was approved in 2002 and was commercially
Stated differently, if Vioxx usage fell significantly before
the recall it would show up in the prescription numbers. It doesn’t. My data
does show a 30% fall from 2003 to 2004. However, that is for the entire year.
You can get similar data by checking [17], This is a “Voice
of the Pharmacist” website. The data is retail-only. It very closely matches the
statistics I have produced so far (ultimately derived from the same sources,
2. Of course, the actual cost of manufacturing brand name
pharmaceuticals is a small fraction of the price (much less true for biologics).
However, the $8.4 billion cited below is not the “cost” of making the free
samples. It’s almost entirely the cost of sending detail agents to doctor
offices. If you check the link, you will see that the $8.4 expense included 116
million detail agent visits to doctor offices. At an average cost of $72.41 per
visit, obviously the money was spent on wages, salaries, and travel expenses,
not manufacturing samples.
3. It is very unclear if Vioxx caused
front-end (the most vulnerable first) or back-end mortality (a cumulative
effect). This is an important issue and contrary indications exist. See
“Q&A: Vioxx’s Health Risks” ( [18]). Merck tried to claim that there was no adverse impact prior to 18
months. See Figure 2 from “Cardiovascular Events Associated with Rofecoxib in a
Colorectal Adenoma Chemoprevention Trial 2005/03/17” ( [19]). If this claim was/is correct, the entire 1999 / Vioxx thesis is
falsified. However, it does not appear to be true (which is why I have avoided
it so far).
Based on additional data and some corrections to the
methods used in the original paper, the NEJM published a correction. See
“Correction – Cardiovascular Events Associated with Rofecoxib in a Colorectal
Adenoma Chemoprevention Trial 2006/07/13” ( [20]). To get a better understanding of the correction, see “Adverse
Cardiovascular Effects of Rofecoxib 2006/07/13” ( [21]). The letters from Nissen and Furberg (noted Merck critics) are
instructive. They reject the 18 month thesis and suggest an essentially linear
cumulative risk model. I quote (from Nissen).
“The original article
included a post hoc hypothesis that curves for confirmed thrombotic events would
not begin to diverge until after 18 months of exposure to rofecoxib. However,
all intention-to-treat analyses in the newly released report show that the event
curves begin to diverge much earlier, generally within four to six months. The
most useful Kaplan–Meier curves, involving intention-to-treat analysis of the
APTC end point, show divergence after only three months of exposure to
At least one author appears to support the 18 month
hypothesis (maybe). See “Time-to-Event Analyses for Long-Term Treatments — The
APPROVe Trial 2006/07/13” ( [22]).
Note that none of the NEJM data shows any hint of
front-loaded risks. The Merck model claims no incremental risk before 18 months.
The contrary analysis seems to show greater risk much sooner. However, the risk
is cumulative and linear. The longer a person took Vioxx the more likely they
were to have some kind of heart failure as a consequence. Stated differently,
they were at greater risk starting soon after they took Vioxx and eventually the
risks added up.
As should be clear, this is deeply problematic for any
effort to blame Vioxx for 1999 mortality. Vioxx was dispensed for 7.33 months in
1999 at a rate of 661,000 prescriptions per month. In 2000, the rate was 1.719
million prescriptions per month. It would appear that far more people took Vioxx
in 2000, than in 1999. However, a person
could try to argue that Vioxx usage accelerated in 1999 and that the full 2000
rate was reached late in 1999. Perhaps. However, let’s assume that it is true.
This means that very few 1999 Vioxx users would have been taking it for 4-6
months (or even 3 months) in 1999. Let’s go further and drop any risk onset
delay and assume a purely linear model (incremental risks start on day one).
That makes the total Vioxx risk equal to the number of prescriptions (which
quadrupled between 1999 and 2000). This falsifies the 1999 Vioxx thesis
However, let’s go further and assume a very front loaded
risk profile (you either die quickly or you don’t die at all) and that Vioxx
prescriptions accelerated to the 2000 rate by the end of 1999. With that
combination of assumptions, the incremental mortality should have been in 1999.
However, it also means that in 2004 there was no one left to die. Clearly Vioxx
wasn’t adding a lot of new users by 2004 (sales had been declining since 2001).
With a front loaded model, the Vioxx recall should not have reduced 2004
4. Another very important question is whether incremental
Vioxx mortality persisted after each person stopped taking Vioxx. Of course,
Merck claimed that the answer was no. However, that may not be true. See “Study:
Health Risk Remains a Year After Quitting Vioxx” ( [23]). Quote “”It was very surprising to me,” says Steven
Nissen, acting chief of cardiovascular medicine at Ohio’s Cleveland Clinic. “I
had always assumed that if you stop taking the drug, the risk would go away.”
Nissen says this data shows that’s not true.
“What it shows us is that you can stop taking Vioxx, and
based upon this study, for the next year, you’re still at increased risk. And,
in fact, the amount of increase is almost exactly the same as we saw during the
three years that people were actually taking the drug,” Nissen
Note also the author (Nissen). Assuming Nissen is correct,
the 2004 Vioxx recall thesis is wrong.
5. As the reader
can see, all of the Vioxx mortality models falsify the 1999 / 2004 Vioxx
thesis. The 18 months to trouble model (See “Vioxx: 18 Months to Trouble? –
[24]) rules out any connection between the introduction of Vioxx in 1999
and any incremental deaths. The front-loaded model (die now or not at all)
allows for a major 1999 impact (with sufficiently accelerated Vioxx adoption in
late 1999), but falsifies the 2004 impact of the recall. The linear cumulative
risk model (apparently preferred by Merck’s critics) falsifies 1999 and 2004.
The persistent risk thesis contradicts the front-loaded model and undermines any
claims related to the recall.
6. The NVSS data and the CDC data show
declines in CVD mortality from 1998 to 1999. The decline may have been less than
other years, but a decline is a decline. If Vioxx was really responsible for
500,000 deaths, the data should have a large spike. No such spike exists. The
smaller decline in 1999 is certainly interesting and may have been related to
COX-2 sales. Note that Celebrex was introduced very early in 1999 and reached
huge volumes in that year (unlike Vioxx). That’s not to say that Celebrex is
responsible for the lesser decline in CVD mortality in 1999. However, it is a
better fit.
7. The CDC noted the upsurge in
mortality in 1999 and analyzed it. See “Deaths: Preliminary Data for 1999”
( [25]). Quote
“The preliminary number of deaths in
the United States for 1999 totaled 2,391,630, an increase of 54,374 from the
1998 total. The crude death rate increased from 864.7 per 100,000 population in
1998 to 877.0 per 100,000 in 1999. The two influenza outbreaks of 1999
contributed to the large increase in the number of deaths (10–12), especially
among the older age groups and for several chronic diseases.”
A later
report (Deaths: Final Data for 2004 – [26]) makes the same point. Quote
“Since 1980, the age-adjusted
death rate has decreased every year except 1983, 1985, 1988, 1993, and 1999.
During these years, influenza outbreaks contributed to increased mortality in
the United States (14,15).”
See the 1998 – 2001 P&I mortality data
online ( [27]) for additional information. Note that the first 1999 mortality
spike was in March (two months before Vioxx was introduced). 2004 deaths were
lower than 2003. Once again, the P&I data provides some insight. Note the
huge spike in late 2003 and the absence of such a spike in 2004. See [28] for a graph. Thank you Peter


#2 Comment By peter_schaeffer On May 14, 2012 @ 11:12 am

The TAC needs to do some fact checking before
publishing something like this. If Vioxx actually resulted in 500,000 premature
deaths it would have shown up in the overall death rate. It didn’t. See
“National Vital Statistics Reports” ( [29]). The overall and age-adjusted death rates fell from 1999 to 2005.
Indeed, the age-adjusted death rate fell faster after 1999 than it did


If the 500,000 statistic was correct, there
should have been at least 100,000 incremental deaths in the peak year from
Vioxx. That’s 33 per 100,000 for the entire U.S. See any blips
in the data of the magnitude? They don’t stand out…


Of course, the incremental deaths
should really show up in the CVD (cardiovascular disease) mortality statistics.
They don’t. See “US Death Rates 1975-2009″ ( [30]). Also see some Arizona specific data (“Trends in Age-Adjusted
Mortality Rates of Deaths due to Cardiovascular Disease, Arizona and US,
1980-2004″ – [31]). The Arizona data is not by itself particularly important (state
level death rate variations are huge). However, the Arizona data exactly tracks
the U.S. overall data.


Is it possible that Vioxx resulted in 50,000
deaths over the period in question? Sure. I don’t have anything approaching the
background to evaluate such a claim. I wouldn’t be surprised either way as to
the truth. For the record, I do have opinions on topics like this. I spent years
deflating Thiomersal / autism claims…


However, there is a larger issue here.
NSAIDs (Celebrex, Vioxx, Bextra, etc.) are all associated with incremental
mortality. Indeed, even Naproxen (also a COX-2 NSAID) has been linked to higher
death rates. However, these drugs are simply too valuable to give up. Ask the
people who take them, if anyone has any
doubts. For many, NSAIDs are the difference between a normal life and ongoing,
severe pain.


This is why the FDA panel voted 31-1
to keep Celebrex on the market. The same panel also voted 17-15 to keep Vioxx
for sale. Even excluding panelists with industry ties, the vote was 8-14
(losing) to approve Vioxx. If Vioxx was really as bad as some allege, why did 8
panelists (with no industry ties) favor its continued sale? Why was the vote in
favor of Celebrex (which is also linked to CVD) almost unanimous? See “10 on FDA
Vioxx panel had ties to companies ” ( [32])


Thank you


Peter Schaeffer


P.S. I have no ties to the drug industry
(other than as a customer). I was once prescribed Naproxen many years ago. It
was astonishingly helpful even though I only took it for a week or two. I have
taken Aleve (OTC Naproxen) from time to time.