The Story of IV-Insulin after MI

The DIGAMI trials teach many lessons--about the failure of good ideas and aggressive therapies as well as the strong draw to spin results

I will show you a glaring example of how even investigators miss (or ignore) obvious uncertainty.

The intervention in question is the aggressive control of blood sugar in patients with diabetes who present with an acute myocardial infarction.

Most know that subcutaneous insulin is a common way to control blood sugar in some patients. But in the hospital, we can be more aggressive and use IV insulin infusions, which allow tight control of glucose levels.

In the 1990s, experts noted that patients with diabetes had higher rates of death after MI. One possible reason for this was an interaction between high blood sugar and cardiac injury. This led to the plausible idea that tighter glucose control in patients early after MI would reduce death.

Swedish investigators put this to the test in the DIGAMI trial. We covered the trial last week over at the CardiologyTrials Substack, with a post and podcast.

DIGAMI randomized about 600 patients with diabetes (and an elevated blood glucose) who presented with acute MI to an insulin infusion for a day or more, then subcutaneous insulin. Control patients received standard care in the CCU.

I will copy what the authors wrote regarding their primary aim (or endpoint).

The primary aim of this study was to evaluate whether the insulin-glucose infusion followed by multidose insulin for 3 months reduced mortality at 3 months in diabetic patients with acute myocardial infarction.

With 3-month death rate as their target, they set out to estimate the number of patients needed. They first estimated the 3-month death rate as 35%. That 1 in 3 patients would die from heart attack seemed like a high estimate, but what comes next is even more curious.

The authors then estimated insulin infusions would reduce death by 30%. They don’t explain this estimate, but only a few years before, in the ISIS 2 trial, thrombolytic therapy with streptokinase—an intervention that can stop a heart attack by opening the artery--only reduced death by 25%.

With those enthusiastic estimates, DIGAMI authors enrolled only 600 patients. (I will come back to this.)

The results

At 3-months, 12.4% of patients died in the insulin group vs 15.6% in the control arm. That 3.2% absolute risk decrease, however, could not be declared statistically significant. The authors don’t give the confidence intervals nor the p-value. They just say the p-value was non-significant.

I put the numbers in a calculator and got a relative risk of 0.80 and 95% confidence intervals of 0.5-1.2. The translation is that the insulin infusion arm could have reduced death by 50% or increased it by 20%.

The authors don’t stop there though. They do two other things. First, they present death rates at one year, which were 18.6% and 26.1% respectively. That p-value was 0.03.

They also present a subgroup analysis, wherein they separate (the already underpowered) main groups into four subgroups based on severity of diabetes and cardiac risk. The lowest risk group (non-insulin-dependent diabetes and low cardiac risk) had the greatest risk reduction from the insulin infusion.

They also reported adverse events. Low blood sugar (hypoglycemia) occurred in 15% vs 0% in the treatment vs control arm.

The Spin

Spin is defined by language that distracts from non-significant primary endpoints. Recall that the primary endpoint was death at 3-months—which did not reach statistical significance because of wide confidence intervals.

DIGAMI authors not only had spin in the conclusions, they even spun the title. Note the picture below where I have highlighted the fact that they emphasized the “positive” non primary endpoint.

The conclusions in a leading cardiology journal also ignored the primary endpoint.

Insulin-glucose infusion followed by a multidose insulin regimen improved long-term prognosis in diabetic patients with acute myocardial infarction.

The Closing Chapter

Ten years later, the same group performed the DIGAMI 2 trial. They randomized slightly more patients, included three groups (2 with IV insulin and one with standard control).

They reported no significant differences in all-cause mortality among the three groups. The confidence intervals were still a bit wide but actually hinted at greater risk in the IV insulin arm. (IV insulin vs control: HR=1.26, CI=0.92–1.72). This time they concluded that early IV insulin therapy did not reduce death. (We will cover DIGAMI 2 in greater detail coming soon on CardiologyTrials.)

Take-Home Lessons

The treatment of high glucose levels was an important question to study. The stress of a heart attack often causes high glucose levels, so it was possible that lowering these levels would help. Only an RCT could answer that question. I laud investigators for doing trials.

But since the results of trials guide practice, it’s crucial to conduct and interpret them properly.

DIGAMI 1 used an inflated rate of death. They estimated that 1 in 3 patients would die at 90 days. The actual death rate was half that at 15%. Then they guessed that insulin infusions would shred death rates by 30%. These outsized estimates resulted in too few patients.

But. Instead of concluding that they could make no reliable conclusion, they changed focus to one-year results.

The “benefit” at one year is likely to be a spurious finding—as there were no differences in fasting glucose levels at one year and DIGAMI 2 found no benefit. The authors also spent little focus on the higher rates of low blood sugar in the treatment arm.

I don’t know how many patients were treated with IV insulin after MI in the 10 years between the DIGMAI trials, but surely it was some. Spin in this case likely led to patients being exposed to a potentially harmful more aggressive intervention that provided no benefit. That is bad.

The other notable feature of this story is that the authors surely had no financial conflicts. The enthusiasm that clouded their ability to recognize uncertainty did not stem from financial gain.

My guess is that these scientists were like many of us in medicine: we want our ideas to be helpful. We all have intellectual biases—which, as this story shows, can be quite strong and problematic.

A third take-home is that aggressive interventions often fail. The history of medicine is replete with examples.

An alternative theory to the idea that high blood sugar after MI is detrimental and needs to be treated aggressively, could be that high blood sugar might be an important adaptation to stress—like sinus tachycardia.

Let me know what you think. There’s plenty of spin in our medical evidence base today. But it was also there many years ago.