Where sailors once looked out on seas full of mystery, and populated by monsters, Scylla, Charybdis, the Sirens and Circe, we mariners of a scientific age prospect untold dangers at the pharmacy: Vioxx, OxyContin, and now it would seem, Avandia.
Some 20 million Americans take Avandia (rosiglitazone), a pill that lowers blood sugar levels for those who have diabetes -- or rather, if recent sales data are any indication, 20 million Americans were taking Avandia until the media reported on a study rushed online by The New England Journal of Medicine (NEJM), which claimed to have discovered a 43 percent risk in the increase of heart attack associated with taking the drug.
The results found by Dr. Steven Nissen and Kathy Wolski were so startling they prompted an accompanying editorial in NEJM by Drs Bruce Psaty and Curt Furberg to conclude that: "Insofar as the findings of Nissen and Wolski represent a valid estimate of the risk of cardiovascular events, rosiglitazone represents a major failure of the drug-use and drug-approval processes in the United States."
Strong wording indeed, and reflective of a scenario that is becoming too familiar: the blockbuster drug as a killer, the data that went ignored, the regulators that were too blind or stupid or corrupt to do their job. No wonder Rep Henry Waxman (D-CA) is hauling the Food and Drug Administration into oversight hearings on June 6.
But as powerful a figure in medicine as Dr. Nissen is (he was an early critic of Vioxx, and derailed another diabetes drug on the eve of its approval), there is no getting around the distinctly awkward fact that the statistical methods used to assess the risk from Avandia are controversial; indeed, they are so controversial that one critic was willing to write the entire study off as being hopelessly flawed, while many experts have not yet found sufficient cause to recommend patients give up the drug.
Unfortunately, in their rush to warn Americans about the next big bad drug threat, most news organizations just didn't stop to analyze what the study actually said; if they had, (which would have required them to read the study first), mass panic might have been averted.
Alarm bell one: Nissen and Wolski used meta analysis to pool data from multiple separate studies in order to try and construct a bigger picture of the risks of heart attack and death among diabetics taking Avandia. Meta analysis is a controversial tool in statistics for a number of reasons (but not least that the criteria used and patients studied in one study may not be exactly comparable to another similar study) and the results of such studies need to be treated with caution.
Alarm bell two: The individual studies were not originally intended to assess heart attack risk or cardiovascular death, as Nissen and Wolski themselves note in their paper. This is a critical limitation because a clinical trial prospectively identifies the endpoints it is trying to evaluate. Data mining a group of studies after the fact for a different variable always carries the risk of throwing up false positives: Avandia is correlated with tooth decay, to give a hypothetical example. But more problematically, we don't know how much at risk for heart attack the patients in these studies were from the get go (arterial plaque does not accumulate over a few weeks).
Alarm bell three: They excluded studies which didn't find any incidence of heart attack, which may have skewed the final correlation (for more on this, see Diabeteshealth.com).
Alarm bell four: None of studies by themselves showed a statistically significant association between heart attack (myocardial infarction) and taking Avandia. Statistical significance was only met by adding all the studies together; and the number of excess incidents in the Avandia group is low enough that, as the authors admit, were the incidents classified in a different way in the original studies, the results could have been totally different.
Nissen also made seemingly contradictory statements to the media, which, at least from a journalistic perspective, raises a modicum of concern.
* "I was physically sick to my stomach, and I couldn't sleep" - his reaction to seeing the study's results as told to the AP
* "It needs to be confirmed with other analyses, but if the data hold up -- and I believe that they will hold up -- it does represent a huge public health concern" - to ABC News
* "The magnitude of the effect we saw is sufficiently disturbing that it seems very unlikely to me that effect is going to go away" - to UPI
* "What patients need to know is there is some evidence of cardiovascular harm from Avandia -- but there is not any reason to panic" - to WebMD (via CBS)
Why give such alarming sound bites to the regular press and a message of calm to the news organization that knows something about medicine? Why, if so concerned about the risks to the public, did Nissen give the results of his study to NEJM before informing the Food and Drug Administration? And given that Waxman announced oversight hearings on Avandia so rapidly after NEJM published the study online, was the study provided to him in advance? Is there some shared agenda here -- and if so, who exactly is putting the public at risk?
And now, to really ratchet up the political conspiracy mongering, Nissen is embroiled in charges that he's running something akin to a protection racket at his clinic, which is funded by, among other companies, the manufacturer of Avandia's one rival in the marketplace, Actos.
The pharmaceutical companies are to our times what the fickle gods were to the Greeks, the dispensers of bewildering fortune and misfortune in random measure. Which is why we turn, or should turn, to statistics to plot a course out of confusion and fear. Nissen and Wolski found a signal, which may -- or may not -- be a signal of some risk. But the weaknesses in the study point us to a compelling subplot too, and that is whether one of the most prominent medical figures in the nation, whether by sense of eminence, or depth of conviction, has decided to take on the FDA, and crush a drug he has long been critical of, with preliminary, fragile and impeachable data. The Greeks have a word for this, too: hubris.