My views on Avandia

Ok, so the news about Avandia has exploded across the blogosphere; so maybe someone could help explain something for me. I know just enough clinical statistics to be dangerous (I’ve had one course).

Concering the Nissen and Wolski NEJM paper, I have to ask: what’s the big fuss? The data is horrible in my opinion. Let me show you what I’m talking about.

You can click the thumbnail below to see the full table, but notice the confidence intervals. The first set is the odds ratio of having a heart attack while on the drug, and every set of data (except the overall combined meta-analysis) shows that there is no difference between rosiglitazone and the control group (and the p-values are really high which means there is a high probability of achieving these results by chance).

Ok, first important thing to note, is that each individual study was 95% confident that the risk of having a heart attack or dying while on rosiglitazone was no greater than their control group (in some cases placebo and in other cases it was medication like metformin or glyburide).

In studies like this, you always have to start with a null hypothesis, otherwise called a “no difference” hypothesis. This is the hypothesis that there is no difference between whatever you are testing, and a control group. The reason they can’t say there is any difference is because their confidence intervals INCLUDE odds ratio = 1. If your odds ratio equals 1, then that means there is not an increase (or decrease) in the odds of developing whatever result you’re watching for (MI or death) while taking the experimental treatment (compared to the control).

THE ONLY result that was significant was the meta-analysis of the insignificant trials. So, I have to wonder (like I did about the acupuncture-blood pressure result mentioned at the end of my last post), even though these results are statistically significant, are they clinically significant? This leads me to their experimental design.

In their Methods section they state that “trials in which patients had no adverse cardiovascular events in either group were excluded from analysis.”

Um, excuse me? WHY?! Am I just completely naive? Isn’t this throwing out data? I realize they are only looking at cardiovascular events and if the studies did not report results for cardiovascular events then obviously they can’t be used; but I think they’re saying something different here. And depending on how many data points they threw out, their results could be totally different! Am I wrong?

Also, they define the control group “as patients receiving any drug regimen other than rosiglitazone.”

Um… great. To my knowledge, only ONE trial (out of the 42 that met their criteria) actually had a placebo as the control. All of the others used comparator drugs. And the study with the placebo wasn’t testing diabetics (like the other groups were). They were testing people “at high risk” of type 2 diabetes. This makes it harder in my opinion to say that these effects (heart attack and death) are due to the experimental variable (rosiglitazone). Obviously, Avandia was approved by the FDA because the data at the time said it was safe. This study certainly doesn’t convince me otherwise.

If I’m going to accept someone’s conclusions, they’re going to have to SELL them to me. Convince me. So far, this paper isn’t convincing. I don’t see why I should accept a poorly designed meta-analysis (with missing data!) over the previous studies that determined Avandia is safe.

Also important to note: in a prospective cohort trial (more powerful than the Avandia case reports) studying a drug in the same TZD (thiazolidinedione) class called pioglitazone (Actos), they determined that pioglitazone actually has a protective effect against heart attack and death AND another study comparing lipid and glycemic effects of these drugs in type 2 diabetics found that both drugs increased cholesterol levels, but Actos raised HDL cholesterol greater than Avandia, whereas Avandia increased the LDL cholesterol higher than Actos (both cited in Nissen and Wolski’s paper). And Actos even lowered the overall triglyceride levels, whereas Avandia raised them.

In any case, the researchers have qualified this study by saying that they did not have access to all of the original data. But still, just because you got a result that was barely significant, and it was really difficult to organize the data into a powerful meta-analysis, you shouldn’t shout to the world that this drug is killing people.

If I’m completely wrong, I’d appreciate someone explaining their rationale to me.

~ by jryanstevens on June 14, 2007.