About this Author
DBL%20Hendrix%20small.png College chemistry, 1983

Derek Lowe The 2002 Model

Dbl%20new%20portrait%20B%26W.png After 10 years of blogging. . .

Derek Lowe, an Arkansan by birth, got his BA from Hendrix College and his PhD in organic chemistry from Duke before spending time in Germany on a Humboldt Fellowship on his post-doc. He's worked for several major pharmaceutical companies since 1989 on drug discovery projects against schizophrenia, Alzheimer's, diabetes, osteoporosis and other diseases. To contact Derek email him directly: Twitter: Dereklowe

Chemistry and Drug Data: Drugbank
Chempedia Lab
Synthetic Pages
Organic Chemistry Portal
Not Voodoo

Chemistry and Pharma Blogs:
Org Prep Daily
The Haystack
A New Merck, Reviewed
Liberal Arts Chemistry
Electron Pusher
All Things Metathesis
C&E News Blogs
Chemiotics II
Chemical Space
Noel O'Blog
In Vivo Blog
Terra Sigilatta
BBSRC/Douglas Kell
Realizations in Biostatistics
ChemSpider Blog
Organic Chem - Education & Industry
Pharma Strategy Blog
No Name No Slogan
Practical Fragments
The Curious Wavefunction
Natural Product Man
Fragment Literature
Chemistry World Blog
Synthetic Nature
Chemistry Blog
Synthesizing Ideas
Eye on FDA
Chemical Forums
Symyx Blog
Sceptical Chymist
Lamentations on Chemistry
Computational Organic Chemistry
Mining Drugs
Henry Rzepa

Science Blogs and News:
Bad Science
The Loom
Uncertain Principles
Fierce Biotech
Blogs for Industry
Omics! Omics!
Young Female Scientist
Notional Slurry
Nobel Intent
SciTech Daily
Science Blog
Gene Expression (I)
Gene Expression (II)
Adventures in Ethics and Science
Transterrestrial Musings
Slashdot Science
Cosmic Variance
Biology News Net

Medical Blogs
DB's Medical Rants
Science-Based Medicine
Respectful Insolence
Diabetes Mine

Economics and Business
Marginal Revolution
The Volokh Conspiracy
Knowledge Problem

Politics / Current Events
Virginia Postrel
Belmont Club
Mickey Kaus

Belles Lettres
Uncouth Reflections
Arts and Letters Daily
In the Pipeline: Don't miss Derek Lowe's excellent commentary on drug discovery and the pharma industry in general at In the Pipeline

In the Pipeline

« Exalted Paper | Main | More Avandia, And More on Marketing »

May 24, 2007

Avandia: Trouble or Not?

Email This Entry

Posted by Derek

Steve Nissen has (once again) made waves with an analysis of cardiovascular risk. This time the subject is Avandia (rosiglitazone), a therapy for diabetes that's the oldest PPAR-gamma drug on the market. A meta-analysis of 42 reported clinical trials of the drug led to the conclusion that rosiglitazone is associated with a statistically significant risk of cardiac events.

The similarities to the Vioxx situation are what have made headlines (and what sent GlaxoSmithKline's stock down about 8% on the day the paper was released). But there are some important differences. Merck's ran into the Vioxx numbers in their own clinical data - the arguing has been whether they recognized the effects earlier (or should have), but it was a specific trial of theirs that led to the statistics that sank the drug. A meta-analysis is a much different beast, since you're trying to fit a large number of different trials, run in different ways for different reasons, into the same framework. Not everyone trusts them, even when the analysis is performed by someone as competent as Nissen, who does mention the limitations of the approach in the paper:

"Our study has important limitations. We pooled the results of a group of trials that were not originally intended to explore cardiovascular outcomes. Most trials did not centrally adjudicate cardiovascular outcomes, and the definitions of myocardial infarction were not available. Many of these trials were small and short-term, resulting in few adverse cardiovascular events or deaths. Accordingly, the confidence intervals for the odds ratios for myocardial infarction and death from cardiovascular causes are wide, resulting in considerable uncertainty about the magnitude of the observed hazard. Furthermore, we did not have access to original source data for any of these trials. Thus, we based the analysis on available data from publicly disclosed summaries of events. The lack of availability of source data did not allow the use of more statistically powerful time-to-event analysis. A meta-analysis is always considered less convincing than a large prospective trial designed to assess the outcome of interest."

And that's what's happening here. A number of people at large diabetes treatment centers aren't ready to buy into a cardiovascular risk for Avandia yet, because they're wary of the statistics. There's a large cardiovascular outcome trial of the drug going on now, which won't wrap up until 2009, but several people seem to want to wait for that as a more definitive answer.

If Nissen's data hold up - and statistically, I'm definitely not up to the task of evaluating his approach - then we might be looking at a Vioxx-like risk level. Out of some 14,000 patients on the drug in the various studies, there were 86 heart attacks in the treatment groups, and 72 in the controls. That comes out to be statistically significant, but (as you can see) the problem is that Type II diabetics have a high background rate of CV problems. Looking at Nissen's Table IV, it also seems clear that most of the significance he's found comes from the pooling of the smaller studies. The larger trials are nowhere near as clear-cut, which makes you wonder if this effect is real or an artifact.

I'm certainly not prepared to say one way or another, and I just hope that the ongoing trial settles the question. It's certainly not unreasonable to imagine a PPAR gamma drug having this side effect, but if this were a strong mechanism-based phenomenon the numbers would surely be stronger. If a risk is confirmed, though, we'll then be faced with a risk-benefit question. Does the glycemic control that Avandia provides lead to enough good outcomes to offset any cardiovascular risk over a large population? If you think getting the current numbers is a tough job, wait until you try to work that one out.

Comments (19) + TrackBacks (0) | Category: Cardiovascular Disease | Clinical Trials | Diabetes and Obesity | Toxicology


1. Insider on May 24, 2007 10:14 AM writes...

It's days as a blockbuster have ended.

Permalink to Comment

2. WC on May 24, 2007 10:28 AM writes...

"We pooled the results of a group of trials that were not originally intended to explore cardiovascular outcomes. Most trials did not centrally adjudicate cardiovascular outcomes, and the definitions of myocardial infarction were not available."

So in other words, wait for the 2009 data.

Permalink to Comment

3. DCRogers on May 24, 2007 11:30 AM writes...

There's an additional twist here, if my understanding is correct. In diabetes, control of blood sugar levels is only a means to an end: to protect (among other things) the heart from damage. In the GSK application for rosiglitazone, the trials showed a positive effect on this biomarker (control of blood sugar levels) and did not measure for the endpoint of interest (fewer heart-related events). [As a side-note, some doctors resisted prescribing it because of reports it raised cholesterol levels.]

I believe that's why there is an ongoing trial to measure just that endpoint [heart related events], as part of a post-approval followup. (Hope there's a good mechanism in place to end the trial early if significance is reached... given the numbers of people taking the drug, we're looking at tens of thousands of deaths if the meta-analysis is correct.)

Permalink to Comment

4. MTK on May 24, 2007 12:46 PM writes...


All clinical trials are subject to a IRB (internal review board) that periodically examine the data to assess progress. Despite the name, these are usually outside experts whose job is to halt a trial should their be good reason. There are lots of reasons why a trial would be halted but one is definitely obvious adverse effects. The other would be if the treatment were so superior to the placebo group that it would be unethical to keep the treatment from the placebo group.

Permalink to Comment

5. Jose on May 24, 2007 12:56 PM writes...

I am always amazed at how minuscule these adverse events are... I really wish coverage in the media would highlight this fact, making some comparisons to alcohol, or other risks people are familiar with.
>Out of some 14,000 patients on the drug in the >various studies, there were 86 heart attacks in the >treatment groups, and 72 in the controls.

Permalink to Comment

6. Morten on May 24, 2007 1:07 PM writes...

"tens of thousands of deaths"? How do you get that number? I know there are 100 mio obese Americans but they can't all be diabetic. How many people are taking the drug?
Ok, I looked at the numbers and I get around 20,000 a year too. If 100 mio people are taking the drug. But I do find it incredibly curious that there were more heart attacks in the control group than in the Avandia group. The control group was only 11,634 people btw Derek. You may want to update your post. Does anyone know why there were different sizes in the cohorts in the heart attack data and the death from heart attack data?

On a related note, how do they design the RECORD post-approval study looking for CV risk? They can't give people placebos or look at an untreated control group... is it a "simple" comparison between patients treated with the various drugs on the market? If some medication is better suited for people with high CV-risk then how do you correct for that?

Permalink to Comment

7. Clark Kent on May 24, 2007 1:08 PM writes...

It is interesting the things that don't make headlines from the paper. Like the odds ratio for death from any cause 1.18, P = 0.24. Its only when you classify deaths from cardiovascular causes that you get a "borderline significance" P = 0.06. He says 48 out of 116 studies met his inclusion criteria, but he only used 42 studies because 6 studies had no cardivascular events at all. I wonder what adding those 6 studies back would do to the "bordeline significance"?

I also find it interesting the companies that have and have not donated money to charity for him.

Takeda and Eli Lilly -- yes
GSK -- no

Remember Vioxx?
Merck -- no
Pfizer -- yes


Permalink to Comment

8. Still Scared of Dinosaurs on May 24, 2007 2:25 PM writes...

WC - unless A trial was designed with a cardiovascular endpoint I would not expect central adjudication of events or explicit definitions of events. The diagnosis of MI is made by the MD in the clinic and, I expect, followed up by the study sponsors (at least if it's an industry-sponsored trial). Furthermore, it's reasonable to expect that a good number of these events were classified as SAEs, which entails an even higher level of scrutiny. While ultimately not as convincing the ongoing trial will be the definition and collection of MI events in the metaanalysis is a pretty small weakness in the study design.

MTK - An IRB wouldn't have the direct authority to shut down a study at sites they don't oversee. They could put up enough of a stink to get the study closed down, probably, but they would be shooting themselves in the foot as far as future research would go and I couldn't foresee that happenning. Their primary responsibility is to the patients at their site(s), so their review, as far as it goes, is usually directed to new ongoing findings that may change the risk profile that the patients were given at the time of informed consent and concerns about the conduct of the personnel at their institusions. It's well known that most IRB's are woefully underresourced to fully oversee all the research at their institutions, and many "central IRB"s are little more than rubber stamps for industry sponsors.

What you are referring to falls more in line with the responsibilities of a Data Review Board or Data Safety Monitoring Board. They boards are created for specific trials and given specific mandates as to what data they are asked to review, what decisions they are empowered to make, and often when they are empowered to make them. The DSMB style, to the extent that names have specific meaning, is often called in when large trials of new agents are run or new trials are run with previously found serious safety concerns. The DRB style is usually tasked to evaluate pre-specified interim analyses in an unblinded manner when the majority of study personnel are to remain blinded. I would be surprised if the ongoing trial discussed here didn't have something like this in place.

Permalink to Comment

10. Anonymous on May 24, 2007 3:12 PM writes...

Like Jose, I would like someone get some percentages to put this into perspective.

For example, this is from the Wikipedia site on torceptrapib: "Each study arm (torcetrapib + atorvastatin vs. atorvastatin alone) had 7500 patients enrolled; 51 deaths were observed in the atorvastatin alone arm, while 82 deaths occurred in the torcetrapib + atorvastatin arm."

So, 60% increase over the control group in deaths is enough to sink a trial. Is 19% enough for Avandia? Can anyone suggest a threshold?

Permalink to Comment

11. hhibob on May 24, 2007 4:18 PM writes...


"But the actual denominator of patients used in finding these numbers was different. The authors found 86/14,371 patients or 0.598% of the patients taking rosiglitazone had a heart attack, while 72/11,634 or 0.619% of people not taking rosiglitazone also had a heart attack ... The authors are using odds ratios, which inflate the odds to sometimes misleading proportions."


"So if you look at the table of included studies, "placebo" is sometimes an inert pill, other times it's insulin, other times it's metformin, metformin plus glyburide, or a sulfonylurea - all medicines used to treat diabetes."

Permalink to Comment

12. david on May 24, 2007 4:57 PM writes...

As an epidemiologist, I am amazed at the attention Nissen's piece has gotten. Aside from the obviously poor refereeing on this paper, the misuse of meta-analysis, the disregard for population-level data, and the off-hand dismissal of the company's full disclosure database, there is the little matter of a 49% increase in risk, which is easily attributable to confounding factors. Equally irritating is that Psaty and Furberg went along with this charade when they know better. There may be bona fide risk present, but this study doesn't show any of it. Let's stick to the facts (precious few that there are in this instance). The simple fact is this paper doesn't show anything. Period.

Permalink to Comment

13. Steven on May 24, 2007 7:02 PM writes...

Could you please explain what an odds ratio is? In the paper and the odds ratio for the treated group was 1.64. In the last table in the paper the odds ratio for insulin was >2 and that for placebo was listed at 1.8. Are these numbers compiled from all trials of diabetic medications were there was an insulin arm? (As an aside I was recently told that the odds ratio of developing lung cancer from smoking was 19 and the odds ratio of developing lung cancer from second hand smoke was ~2.) Given the odds ratios in the last table, isn't a 1.64 odds ratio in line with what is seen for the other commonly used medications for diabetes? Also why didn't Nissen do the analysis of the other ppar medications on the market? Does he not have access to the information of those trials?

Permalink to Comment

14. biohombre on May 24, 2007 10:29 PM writes...

But as inferred by DCROGERS, isn't part of what is at issue here is related to use of biomarker? (Not what the national press has made of it, though!) Heart disease is a complication of diabetes. The Diabetes Control and Complications Trial (DCCT) provided evidence that good blood glucose control (lowered blood glucose, glycated hemoglobin) corresponded to lowered risk of complications. Avandia lowers blood glucose (a sort of biomarker for diabetes complications), this should infer that risk of complications, such as heart disease, is lowered. Has anyone seen any indication in the data that risk of heart complications is lowered?

Permalink to Comment

15. Petros on May 25, 2007 2:43 AM writes...

Now will someone do a similar meta-analysis for Actos(pioglitazone) the only other approved agent in this class?

Sales of the two products are fairly comparable so any distinction in risk between the two ought to be discernible.

Permalink to Comment

16. DiabetesDoc on May 25, 2007 11:07 AM writes...

People have looked at Pioglitazone and vascular events - the PROactive trial - showed a 10% decrease with PIOGLITAZONE in the secondary endpoint of all cause mortality, MI, stroke.

Nissen's paper briefly refers to it in the discussion

Permalink to Comment

17. Jose on May 25, 2007 12:03 PM writes...

David- could you say a little more about the relevance of a meta-analysis, and why it is a poor study for this system? We've *far* exceeded my grasp of stats on this one.... Thanks.

Permalink to Comment

18. Roy M. Poses MD on May 25, 2007 1:12 PM writes...

One big issue, IMHO, is that rosiglitazone has never been shown to be superior to any other glucose lowering drug in terms of clinical benefit, i.e., in terms of making people feel better, avoid specific complications of diabetes, function better, or live longer. In the absence of any such benefit, even suspicion of more adverse clinical events would suggest patients would be better off on other drugs whenever possible.

So, sure, the meta-analysis is not definitive (although the DREAM trial did show numeric but not significant increases in adverse cardiac events for patients who got rosiglitazone). But in the absence of a clear benefit, even a suspicion of increased harm is important.

Note further that the DCCT trial did NOT include patients with type 2 diabetes, only type 1 (insulin dependent). Its results are IRRELEVANT here. Type 1 and type 2 diabetes are different diseases.

Note further that the only trial to assess the clinical effects of intensive efforts at glucose control in type 2 diabetes, the UKPDS study, did show that intensive treatment reduced microvascular complications of diabetes (mainly eye complications), but that it also increased the rate of hypoglycemia, including major hypoglycemic episodes. So whether the benefits of intensive treatment of type 2 diabetes with oral agents outweigh the potential harms is also debatable.

And by the way, as far as I can tell, the GSK trials data-base was only posted on the web because GSK was forced to post it as part of a settlement of a lawsuit (which involved allegations that GSK had suppressed clinical research data in other instances.) So while I commend GSK for making this data available, the company did certainly not voluneer to do so.

Finally, I would suggest that this debate suggests that people interested in pharmaceuticals (from any angle) may be interested in learning more about clinical epidemiolgy and the principles of evidence-based medicine. See:
Guyatt G, Rennie D, eds. Users Guides to the Medical Literature.

Permalink to Comment

19. Government Accountability Project on June 19, 2007 3:17 PM writes...

Please pass this on to friends and colleagues interested in issues of drug safety and reform of FDA drug approval processes.

THE GOVERNMENT ACCOUNTABILITY PROJECT (GAP) invites you to join us for “From Vioxx to Avandia: Whistleblowers and Dangerous Drugs” - a Live Telephone Conference Call with Q & A on Thursday, June 21st, at 6:00pm - 7:00pm eastern.

The call will feature David Ross, Ketek FDA whistleblower and Mark Cohen, GAP Food & Drug Safety Director.

To take part: email To receive the call-in info by text message, please include the number in your email.

GAP has been involved in a number of prominent cases related to restricting the presence of dangerous drugs on the market, most notably Vioxx.

The Vioxx scandal came to light when Dr David Graham, a top FDA scientist and GAP client, testified before a Congressional committee that Vioxx had likely caused as many American fatalities (heart attacks and strokes) as had the Vietnam war.

Moreover, as it was structured, FDA could not be relied upon to protect the American public from future drug disasters.

Join us for a live conference call with David Ross, the former FDA medical officer who blew the whistle on the mishandling of the approval process for the antibiotic Ketek, and Mark Cohen, GAP food and drug safety director.

The revelations that Dr. Ross and others made resulted in significant restrictions in the approved uses for Ketek.

The recent controversy over the diabetes drug Avandia seems to confirm Dr Graham’s critique of the agency’s culture and processes.

Reforms being discussed by Congress right now could go some way toward changing the way the agency balances the interests of rapidly bringing new drugs to market and the interests of ensuring that news drugs are truly safe and effective.

Join us on this live conference call to get the latest update on the progress of these reforms and to put questions directly to David Ross and Mark Cohen regarding our work on Drug Safety and the experience of being a whistleblower within the FDA
For call-in details, email

Permalink to Comment


Remember Me?


Email this entry to:

Your email address:

Message (optional):

The Last Post
The GSK Layoffs Continue, By Proxy
The Move is Nigh
Another Alzheimer's IPO
Cutbacks at C&E News
Sanofi Pays to Get Back Into Oncology
An Irresponsible Statement About Curing Cancer
Oliver Sacks on Turning Back to Chemistry