Avandia: Burn Her Anyway?
Three years ago Dr. Steven Nissen published "Effect of Rosiglitazone on the Risk of Myocardial Infarction and Death from Cardiovascular Causes" in which he reported that those taking Rosiglitazone (Avandia) had about a 40% increased risk of acute myocardial infarction (heart attack). The study was a meta-analysis of published and unpublished data and drew extensively from the manufacturer's unpublished data.
The media simultaneously reported the findings and propagated the following narrative: The pharmaceutical companies collect vast quantities of data; publish only what supports their products; and are at best willfully ignorant of the risks of their products which risks are right before their eyes if only they would look. Dr. Nissen who "has a statistician’s zeal for drilling deep into clinical data, seeking signs that some widely used drugs pose undisclosed risks to patients" was made the hero of the drama. Congress got involved by beating the drums of safety and transparency and demanding that the FDA pay more attention to ensuring that pharmaceuticals are safe.
The FDA reassessed Avandia and panel members voted 21-3 to keep it on the market.
In February of this year, a Congressional investigation conferred near-martyr status on Dr. Nissen after it was revealed that he had secretly tape recorded a meeting with the representatives of the maker of Avandia. The same month, in European Heart Journal, he published the editorial "The Rise and Fall of Rosiglitazone". Meanwhile everyone was waiting for the results of a huge study of Avandia.
Last week that study, "Risk of Acute Myocardial Infarction, Stroke, Heart Failure, and Death in Elderly Medicare Patients Treated with Rosiglitazone or Pioglitazone" (interestingly, for several days JAMA made you click through an ad for a competitor's type 2 diabetes product to get to the article) was published. And what was the risk of heart attack among those elderly patients on Avandia? Essentially no different than those on the older medication - a statistically insignificant 1.06 increase - especially so given the fact that almost three quarters of all type 2 diabetics wind up dying of heart disease in any event. So, was Avandia cleared? Nope. In fact, the calls for removing it from the market have grown even louder.
The new study, while rejecting the original claim that Avandia causes heart attacks, raised the hypotheses that Avandia causes congestive heart failure and stroke and increases the rate of mortality overall. Yet even though the numbers of patients involved in the study (their records were simply culled from Medicare databases) was large (227,000) the increases were relatively small and by the time the increases appeared (after six to fifteen months of medication) only a small and rapidly decreasing fraction of the original cohort were actually still on one of the two medications.
At almost the same time Dr. Nissen published an updated meta-analysis ("Rosiglitazone Revisited: An Updated Meta-Analysis of Risk for Myocardial Infarction and Cardiovascular Mortality" also free and also, though published in the Archives of Internal Medicine, only after an ad for the same competitor's type 2 diabetes product) that purports to show that Avandia increases the risk of heart attack but doesn't increase the risk of mortality. What? Hey, wait a second! How can ... So, siding with those advocating "safety over certainty" the New York Times quickly editorialized in favor of those demanding Avandia be removed from the market.
The LATimes went so far as to publish a piece calling for "an immediate moratorium on sales as soon as a credible study raises questions about safety." In that same piece the journalist was posed the following question by a researcher being interviewed: "Suppose a drug saves five people and kills one person. Do you keep it on the market?" His answer was "I know this: If that one person killed is my loved one - or yours - the answer is readily apparent."
Yikes. Really? If a data dredge shows any risk of a fatal outcome then the drug should be pulled from the market no matter how many people are killed in the process? We'll save for another day the question of how thinking about risk goes so badly astray. For now though consider how likely it is (yes likely - and in fact, if the number of endpoints examined are sufficiently large, how almost certain it is) that a risk will be found where none exists. Start with "Data Dredging, Bias, or Confounding: They Can All Get You Into the BMJ and the Friday Papers" and then for more on the perils of statistical deep drilling read "Your Intuitions Are Not Magic" and the links therein.
At the end of the day the issue isn't safety versus certainty. Claiming that pharmaceutical manufacturers are insisting upon certainty before warnings are issued or products are pulled is just a straw man argument. The usefulness of statistical analyses of medical outcomes has not been added to "death and taxes".The real question is which course is less uncertain - making vital judgments on the basis of large randomized controlled trials like the one due out on Avandia in 2015 or on the basis of data dredges? The answer ought to be obvious.
