Articles

Medical Science and Public Opinion: The Avandia Story

The week of 12-16 July saw an FDA Advisory Panel meet to decide the fate of an important drug. Along the way, the FDA charted new territory in using drug comparisons to judge safety, amidst external allegations of corporate malfeasance and patient harm.

Avandia, or rosiglitazone, is one of a new class of anti-diabetes drugs approved for marketing by the FDA in 1999. It, and its competitor, Actos (pioglitazone), are thiazolidinediones (TZD’s), a class of drugs that act to decrease insulin resistance. A third TZD, troglitazone, was withdrawn after studies showed a 1:20,000 incidence of hepatitis and potential liver failure.

Avandia was a clear market success, with sales peaking at $2.5B in 2006, the year prior to the first research “shot across the bow” regarding patient safety.

Among scientists, the mid-July review capped a rising level of concern. Among the public, a tide of safety concern had been rising for several years, flowing from the scientific community into the legal and political arena.

The story is a paradigm of how science and public opinion interact to determine the fate of a medical treatment where there is little dissent over effectiveness, but little agreement over safety.

Setting the Stage for the Review Process

The FDA charges Advisory Panels with evaluating therapeutic agents and devices on the scientific evidence, driven by concerns over two criteria: effectiveness and safety. The panel’s safety assessment must look at risk (side effects, adverse reactions, and dosage limits) in the context of expected benefit.

But with marketing approval by the FDA, questions of risk may still remain. Most companies will try to anticipate safety concerns that arise during the pre-marketing development of the drug. Research protocols are altered to include new findings….but you can’t check for everything. When the FDA judges that the unanswered questions are not of sufficient concern to postpone approval to market the drug, they will direct “post-marketing” studies to fill in the blanks.

Most of the hoopla from these panels comes from disagreements about safety. Drugs and devices that are clearly ineffective never make it to an initial panel hearing. Even if the agent is only slightly or partially effective, this measure is considered met.

Panels do not carry out comparisons between drugs that address the same health benefit to find the best one. With healthcare reform, a closer look at which agents do a better job will become more common, a process known as “comparative effectiveness.” But FDA panels are not charged with that. They compare drugs, yes, but mainly in pursuit of safety and effectiveness, not whether it’s the best.

A vote is taken. The results of the vote constitute the recommendation of the panel to the FDA, who makes the final decision.

Last week, when all was said and done, the panel recommended continued approval for Avandia, but with stronger restrictions on its use, and additional warnings about risk….I think. The voting process–a multiple choice method–left some lay observers in doubt.

From the public’s point of view, these decisions can seem arbitrary, or influenced by outside interests: patient safety advocacy groups or the pharmaceutical industry. Journalistic interpretations of FDA action make no attempt to assess data, so it’s not surprising that many in the country think that science is a process of dueling research studies, a fight to see who can amass the most—and loudest–supporting authority.

But the process is not like that.

The Science Of the Review

The 2-day meeting illustrated the complexity and limitations of determinations of safety. Oral agents in the treatment of diabetes have had a checkered history. Most have significant warnings and precautions, and well-defined–if rare–serious adverse reactions. The issue with Avandia is whether the drug is a significant risk of heart attacks (MI) and other heart disease, and if so, of enough risk that it should be withdrawn.

Two questions needed answering: does Avandia add to the heart risk of diabetes, and if so, is it at odds with that of similar approved oral agents for diabetes?

The advisory committee of the Division of Metabolism and Endocrinology Products had previously met in 2007 to consider just these issues. A signal study combining data from 42 clinical trials had been published demonstrating an odds-ratio of 1.4 (CI: 1.1-1.8, p=0.02) for “non-serious and serious myocardial ischemia.” Back then the committee thought the study was strong enough to be concerning, but there was too much variation in (1) the length of the studies (mostly short), (2) what other medicines the patients might have been taking, and (3) other characteristics of the patient populations. When the committee tried to eliminate this variation (by analyzing sub-sets of the data to control for variation) the odds ratio for ischemia fell to 1.0 (no increased risk) or less.

An even bigger blow came when Avandia was compared to placebo versus Avandia compared to other anti-diabetic agents: the increased risk was present ONLY in the placebo studies. Maybe ALL the anti-diabetic drugs examined were similarly risky.

There were 3 longer term studies available (DREAM, ADOPT, and, preliminarily, RECORD) that showed higher event rates for MI with Avandia, but lower mortality than controls. Not very helpful.

The Office of New Drugs (OND) thought Avandia should be “thumbs up” with warnings, the Office of Surveillance and Epidemiology (OSE) thought it should be “thumbs down.” The findings were submitted to the Oversight Board, which went with OND. Ultimately, not everybody agreed, but the 2007 data was inconclusive, the WARNINGS and PRECAUTIONS label was updated, post-marketing studies were directed (including RECORD, which was ongoing), and Avandia was retained.

Then Why Did They Meet Again In 2010?

Science is an iterative process. You keep going back to look at issues with new variables, under different conditions, and in the light of new data. Particularly if there are continuing reports of adverse events.

Specifically, there was new data comparing Avandia to Actos, and updated information on heart risk. The controversy had also spilled over into the public arena, including lawsuits, and a Senate special report.

Patients, Public Opinion, and Politics Enter the Process

Over 13K lawsuits have been filed against Avandia, claiming harm due to the drug. Potential awards from those suits total between $1 and $6B. Patients clearly noticed there was dissension in the scientific community over safety. This is always a spur to litigation. In the Vioxx case, lawsuits exploded similarly. By May of 2010, the company GlaxoSmithKline (GSK) had agreed to settle 700 Avandia suits for a total of $60M.

Constituent complaints had pushed the Senate Finance Committee to conduct a two-year investigation. The report, released in February of this year, alleged the company (GSK) was aware of the possible cardiac risks years before such evidence became public. It also charged that the company intimidated physicians, focused on strategies to minimize findings that Avandia may increase cardiovascular (CV) risk, and sought ways to downplay findings that the rival drug Actos was safer.

Politics, trolling trial lawyers, and public opinion were sniping at the ‘07 review, and pressing for a new FDA review. The Director of OSE, who had argued in 07 to withdraw the drug, sent a memo in Oct of ’09 up the chain urging reconsideration:

“The benefits of rosiglitazone do not outweigh its risks. I recommend granting the Citizen Petition’s request that FDA remove rosiglitazone from the market.”

The FDA decided to re-open the question of Avandia’s safety; directives were issued to gather data for a hearing in July.

The Science In 2010

Background material alone for the 2010 Advisory Panel meeting was 765 pages long. Sessions from both days were Live-Blogged for internet viewers (a large number of whom were traders attempting to anticipate stock prices).

New data included completion of the third long-term study (RECORD), comparing Avandia to older anti-diabetic drugs. The data showed that Avandia was no more risky than the others with regard to cardiovascular death. The manufacturer argued that this supported removing the (CV) warning from the label. The study, conducted by the manufacturer, was criticized on several grounds by the Division of Cardiovascular and Renal Products, (DCRP) with allegations of “mishandling” of cases (recording errors, record changes, improper adjudication of endpoints).

An FDA cardiology consultant from OND (recall that office had recommended in ‘07 to keep Avandia) reviewed the DCRP allegations, and while admitting some limitations, argued in just as much detail that the DCRP erred by re-defining the study’s endpoints (post-study) to include more CV events. Admitting there was “ascertainment bias” in RECORD, he pointed out that even if you corrected for the alleged bias, the odds ratio was still not statistically significant for increased MI risk. The ratios were as follows:

RECORD  odds ratio 1.14 (CI: 0.80-1.63; p=0.47)
DCRP  odds ratio 1.38 (CI: 0.99-1.93; p=NS)

Further he subjected the original 2007 study to withering criticism, concluding that (1) the studies included were so small and short-term that slight changes in the data could have had big impact on the odds ratios, (2) they weren’t adjudicated, (3) they weren’t originally designed to look at CV risk, and (4) they didn’t define the signal event of MI.

The lead investigator for the 2007 study admitted the weaknesses, but insisted that even so, after 10 years GSK should have stepped up to better studies (according to the Live Blogging record) The result was to make the original study look like more smoke than fire…if still a concern.

In sum, there were unfavorable trends for MI in both analyses but neither was statistically significant. The cardiologist then added a hammer blow that when you lumped MI in with all-cause mortality, risk disappeared: In his words,

“results on the primary endpoint (time to first cardiovascular death or cardiovascular hospitalization) are not entirely conclusive. Conversely, the findings on all-cause mortality seem readily interpretable and important in RECORD, and they favor rosiglitazone.“

It was déjà vu all over again. With this analysis, the consultant had undermined the original signal 2007 study and minimized criticism of the long-term study. The panel then shifted its attention to the TZD alternative, Actos: would it be a safer alternative? In the minds of some, this would be sufficient argument for withdrawing Avandia.

One panel member asked, in effect, “Are we creating a precedent by comparing drugs in this way?” The answer given was, “Yes, but not in such a way that implies one drug is better than another.” This answer puzzled some observers, but having treatment choices available is a strong argument for preserving alternative agents for those patients who might benefit from one drug but not the other. This bordered on exceeding the FDA’s authority. As I said earlier, they do not do “comparative effectiveness”. They focus on an individual drug’s efficacy, bounded by safety.

The data comparing Avandia and Actos, when presented, were confusing and complex. The observational studies varied so widely in design that the statistician could come to no firm conclusion. By contrast the meta-analysis by OSE (the thumbs down people) concluded

Comparisons of rosiglitazone with pioglitazone consistently show a clinically meaningful increased risk of adverse cardiovascular outcomes, especially acute myocardial infarction, with rosiglitazone.

But what about the argument over alternatives? Does a suspected increased risk (note they said “clinically meaningful” increase rather than “statistically significant”) warrant removing an alternative? If so, what do patients who can’t take Actos do?

In summary, all the studies had biases and limitations. But, despite all the lack of certainty, there’s a clear warning: the heart risk may be there, underneath it all.

The Vote and It’s…Result

The vote was set up as follows. Each member would vote for one option:

  1. Allow continued marketing and revise the current label to remove the boxed warning and other warnings regarding an increased risk of ischemic CV events, or
  2. Allow continued marketing and make no changes to the current label, or
  3. Allow continued marketing and revise the current label to add additional warnings
  4. Allow continued marketing, revise the current label to add additional warnings, and add additional restrictions on use)
  5. Withdrawal from the U.S. market

The results were:

  1. 0
  2. 3
  3. 7
  4. 10
  5. 12

That a plurality voted to withdraw, but the majority wanted to allow continued marketing in some form was a head-scratcher to some. But clearly, had the choices been withdraw or not withdraw, the results would have been to “not” withdraw (By comparison the vote in 2007 was 22 to 1 to continue marketing Avandia). Nobody wanted to allow the manufacturer to remove the warning label regarding CV risk (as GSK requested). Clearly, the weight of opinion from the scientific community was shifting, in parallel with public opinion.

Historically, the FDA decision-makers rarely dissent from panel recommendations.

The Public Aftermath

Law firms were quick off the mark (same day) to inform clients and the general public that the recommendation to allow Avandia to continue didn’t preclude them from suing based on the company’s alleged malfeasance in hiding unfavorable data on Avandia. The manufacturer was also quick to issue a press release with the voting results, re-iterating its position that with suitable precautions, Avandia does not increase CV risk. The media produced a flurry of articles, that all seemed to say slightly different things. Their descriptions of the result were: Avandia “dead,” “can stay,” “a zombie,” “kept,” “revealed mixed opinions,” “restricted.” All over the map.

Sales of Avandia have been dropping since 2007. Given public awareness of the controversy, why would a patient prefer it?

Ominously, the week after the panel meeting the FDA ordered a halt to enrolling patients in a GSK study (TIDE) that would directly compare Avandia to Actos, citing the need to come to grips with Avandia’s safety before deciding on further research.

There’s already more data than usual. The panel’s review did not achieve closure. No one knows whether longer term studies will confirm the claim of risk. My guess: Avandia is on the way out. The momentum of the science, and public opinion, is in the direction of withdrawing the drug. Withdrawal will probably occur before the science is settled. But, no matter what happens to the drug, the legal actions will go on for years.

Vioxx was withdrawn in 2004. The most recent Vioxx law suit against Merck occurred in March of 2010.

About the Author

James Dougherty, MD, MPH, Brig Gen USAF (ret), is a Fellow of the American College of Preventive Medicine. His 35 years in medicine have been split evenly between clinical practice, from F-16 flight surgeon to consultant to the AF Surgeon General, and medical management, from regional health care planner to CEO of a medical center. He developed the first backpack-based surgical team for Special Operations, defended the anthrax vaccine in testimony before Congress, and rebuilt a medical center destroyed by Hurricane Katrina. He was briefly an instructor in philosophy at the University of Md. He has a continuing interest in (1) data-driven decision making in medicine and public health policy, and (2) medical ethics.

Posted in: Pharmaceuticals, Politics and Regulation

Leave a Comment (9) ↓

9 thoughts on “Medical Science and Public Opinion: The Avandia Story

  1. passionlessDrone says:

    Hi James Dougherty –

    Very nicely done.

    - pD

  2. JayHawkDoc says:

    Great article! It is very interesting to see such complex issues laid out in such a clear and linear presentation.

  3. Dr. Le Petomane says:

    This is well-written and very useful. I love the posts on this blog, but this one is particularly applicable to my everyday practice. More, please!

  4. rork says:

    I much appreciated this article, but as a missionary statistician looking for teachable moments will nit-pick that if you give a confidence interval, you should say what confidence it has (I guess 95%), that “p=NS” is not a p-value (maybe it was .054 there, eh?), and that having some idea of the sample sizes is important. The days of .05 being magic are hopefully behind us, the question always being “how strong was that evidence”.

  5. rork,
    Your point’s well taken. It bothered me, too. Unfortunately, this is how the cardiologist published his analysis, so I reported it as written. Thanks.

  6. cervantes says:

    I think Rork’s point requires more than such an airy dismissal. The concept of statistical significance is widely misunderstood and misused, particularly by drug companies. When you have several findings, all in the same direction, but no individual study happens to reach the arbitrary p<.05 level, they will say "there is no relationship . . . " or "no significant relationship . . . " But that is not true. Even with a single study with a p value of .054 or so, it is highly probable that a relationship exists, particularly if you have plausibility and some prior reason for forming the hypothesis. (Bayes theorem is extremely important in interpreting p values but usually ignored.) When you have multiple studies all with a similar trend, you should start to be convinced. When safety is at stake, the precautionary principle also argues that we should be lenient about p values in the first place.

    Let me also say that while diabetes can have many unpleasant complications, the main way it kills people is through CV disease. So a treatment that increases CV risk, arguably, is utterly pointless unless a patient explicitly agrees to trade off life expectancy and risk of disabling heart disease for a lower risk of other disabilities. I seriously doubt many patients have that conversation with their doctors.

  7. cervantes,
    Other than the airy part, I agree with you completely. The p-value wasn’t available.

    And arbitary p-value thresholds are not the whole story, as you both point out.

  8. The Blind Watchmaker says:

    Actos will go generic soon. Avandia will go away anyway.

  9. nitpicking says:

    Speaking as a diabetic who actually takes Avandia, I can’t say I’m terribly frightened of CV disease. At most, there’s a barely-measurable increase in risk.

Comments are closed.