Wednesday, March 28, 2012

The IMMEDIATE trial: Should EMS give Glucose-Insulin-Potassium?

The results of the IMMEDIATE trial have been popping up repeatedly today on Facebook, partly because I "like" a few EMS FB pages, and also because one of the authors (Hi Carin!) is a FB friend and recent Yale ED attending.

Here's an example of the way the trial is being described:


"ACS patients benefit;" "cut the risk of death in half." Sounds great! I love medical reporting/press releases. No pesky nuance or qualification. Me no need anyhow.

The result they are describing, to be specific, is that 8.7% of the people getting the placebo had a cardiac arrest, or died while they were hospitalized, while only 4.4% of the patients getting the study drug did. That's either an (absolute) difference of 4.3%, or about a (relative) 50% decline.

Such an effect would be stunning.  In the years after thrombolytics and aspirin were introduced, the incremental benefits of new therapies for AMI have been getting smaller and smaller. This result here would blow the others out of the water.

For instance, back in 1988, it was shown in ISIS-2 that either the use of aspirin or of thrombolytics reduced the risk of death in MI by about 2-3% over placebo. The combination was better of course.

After that, it's been harder to show that the more complicated and expensive therapies save that many more lives. When we send a patient to the cath lab for an AMI (instead of giving a thrombolytic in the ED), for example, there isn't that huge a benefit. One recent analysis suggested that, overall, you could only find a 0.7% difference in mortality (6.6% vs 5.9%) between lysed patients, and those that went for PCI. A lot of money for not much gain.

So, if this combination of glucose, insulin, and potassium (GIK) could cut mortality in AMI from 6.6% to, say, 3.3%, it would be freakin' amazing.

"I bet there's a catch. There's always a catch."
Well, I don't mean to be an Eeyore, but the perhaps we should wait for, yes, "further study." I offer three reasons why:

1. They weren't studying mortality.

The principle outcome they were studying was whether the initial presentation of ACS would progress to an MI, or it would be an "aborted" MI. This is the outcome that they believed had the most biochemical and clinical justification, and they clearly thought that it had a reasonable chance of being demonstrated.

It turns out there was no difference in the percent of people who progressed to completed MI - the GIK infusion did not help, at least not here. So the trial is negative for the real primary outcome.

2. There were 12 secondary outcomes.

Look at the table of the results:



Remember: the outcome they staked the success of the trial on was the one at the top: "Progression to MI," for all participants.  The rest are a bunch of secondary outcomes, and they don't count to the same degree as the primary outcome.

Analogy: A friend is flipping a coin, and you call heads. That's your primary outcome of interest. But if you also say to your friend "Okay, I call heads, but I also call it if you drop the coin, if it flips over 5 times in the air, if your phone rings in the next 30 seconds, or if your nose starts to itch in the next 10 seconds.

Now, you may be wrong about heads, but say your friend's nose does indeed start to itch in the next 10 seconds? Will he concede defeat? What will he say?

"No pick! NO PICK!" 
Most likely your friend will point out that the most relevant and important prediction you made was heads vs tails. Furthermore, you called out such a long list of other items that you were almost certain to come up with a positive result. He will insistent on another coin toss, where the primary outcome is now nose-itching, not heads or tails.

The same holds in statistics and study design, and is also why the authors state in their conclusion (my emphasis):
"The primary end point was not significantly different between groups, and the observed favorable results of GIK were based on prespecified but secondary end points, although biologically plausible and consistent with preclinical studies. The study tested one primary hypothesis, 3 major secondary, and 6 other secondary hypotheses. All were prespecified and no adjustment for multiple comparisons among the secondary end points was made; thus, reported significance levels should be considered approximate. Accordingly, given the lack of complete consistency of the findings, and the modest P values for most of the statistically significant findings, it would be appropriate to describe the observed favorable effects on the secondary outcomes as generating clinically testable hypotheses for evaluation in larger cohorts."

3. 30 day mortality seems pretty important too...

Ok, say you can take the "cardiac arrest or in-hospital mortality" results at face value. What, then, shall we make of the 30-day mortality? It was shown to be basically the same in both groups.

We just saw this discussion take place last month. A study from Japan showed that giving epinephrine in cardiac arrest got people to the hospital with ROSC more often, but the 30-day mortality was no different (We'll leave the neuro results alone for now.).

It would be nice if epi put all the dots on the right side of the graph. But it doesn't.
So, say the results are right - people don't die or arrest in the hospital as often, but they still die in the first 30 days just as often. Now, maybe everyone's hospital stay was over 30 days, but I doubt it.

Still feel excited?

Bottom line:

If they conduct another study that confirms the mortality benefit, it would be the greatest thing since the free coffee machine in the ED break room. But, unlike the coffee machine, such results are conjecture for now.

No comments:

Post a Comment