The Quandary of a "Positive" Trial with a Non-significant Result?
The MINT trial tested two strategies of dealing with a common problem. The results seem clear. Except for the statistical test.
Don’t get hung up on the specific medical problem. The point of this Study of the Week is how to think about trial results.
The medical question comes up pretty often:
What to do with moderate anemia (low red blood cell counts) in a patient recovering from a heart attack (myocardial infarction = MI)? (We measure something called the hemoglobin level or Hgb to quantify the anemia.)
The two answers are a) be liberal and transfuse blood, or b) be restrictive and withhold blood unless the Hgb level gets really low.
The MINT trial tested these two strategies. No drug companies. No devices. Just strategies. I love it!
The background
After a heart attack, the heart muscle needs to recover. In favor of transfusing is the argument that having enough red blood cells may help oxygen delivery and limit damage. But, on the other hand, giving blood comes with the risk of volume overload, infection, clotting and inflammation.
Previous trials of treating anemia in post-MI patients have been small and inconclusive. But, in general, blood transfusion strategies have been studied in many other clinical situations. A Cochrane review of trials of more than 21K patients have shown that restrictive strategies have led to a 50% decreased use of blood without differences in morbidity or mortality.
In the hospital, restrictive transfusion policies are in vogue.
Some Details of MINT
The restrictive strategy reserved transfusion for Hgb = 7-8 g/dl vs the liberal strategy that allowed transfusion for less than Hgb 10 g/dl.
The primary outcome was strong—MI or death at 30 days. About 3500 patients were randomized equally to either strategy.
Patients were age 72 years; they had normal heart function and the median Hgb = 8.6 g/dl.
The results:
The mean number of red-cell units that were transfused was 0.7 in the restrictive-strategy group and 2.5 in the liberal-strategy group. That makes sense.
This resulted in a Hgb level that was 1.3 to 1.6 g/dl lower in the restrictive-strategy group than in the liberal-strategy group.
Death or MI, the primary outcome, occurred in 16.9% of patients in the restrictive strategy vs 14.5% of patients in the liberal strategy.
Notice there are no hazard ratios in the main figure. I am not sure why.
To my eyes, and probably yours too, the the liberal strategy looked better.
The absolute risk reduction was 2.4%. And. This was a strong endpoint.
The Main Question
Was this significant? That is the hard part.
The crude risk ratio for restrictive vs liberal was 1.16 or 16% worse. The 95% confidence intervals (CI) ranged from 1.00 to 1.35. We usually say that if 1.00 or no difference was included it is not statistically significant. But it gets slightly more complicated.
There was incomplete follow-up on 20 patients in the restrictive group vs 37 patients in the liberal strategy. group.
Now, with adjustment techniques called multiple imputations, the estimated risk ratio came out to 1.15 with CI to 0.99-1.7 and a p-value of 0.07. Darn it. The p-value is now greater than the threshold of 0.05.
But let’s look at some other endpoints.
Death was 19% higher in the restrictive arm. MI was also 19% higher in the restrictive arm.
Here is another endpoint: Death due to cardiac disease was 74% higher in the restrictive arm. The CI went from 1.26 to 2.4.
The frequency of heart failure and other safety-outcome events were similar in the two transfusion groups.
The conclusion in the New England Journal of Medicine has a hedge. (Italics)
In patients with acute MI and anemia, a liberal transfusion strategy did not significantly reduce the risk of recurrent myocardial infarction or death at 30 days. However, potential harms of a restrictive transfusion strategy cannot be excluded.
Comments
I like this conclusion and this trial. It makes you think.
The last phrase adds nuance to the interpretation. I think common sense and strict statistical thresholds (p= 0.05) come into tension.
I am biased about blood. I think in the setting of an MI, when you want to limit myocardial injury, it makes sense to have enough red corpuscles delivering nutrients.
But you could come back and say, come on Mandrola, you are the evidence person, there are oodles of studies that find that restrictive strategies of blood transfusion save blood products without affecting outcomes. And this is a negative result!
I may be wrong, but I think we need to look at this trial with more common sense than statistical testing.
There were more primary outcome events in the restrictive strategy. 2.4% is not a small increase in risk in a cardiac trial.
The 95% confidence intervals for the primary outcome mostly include worse outcomes: 0.99-1.34. That translates to the restrictive strategy being 1% better to 34% worse. The bulk of that interval is worse.
Each component of the primary outcome (MI and death) were higher in the restrictive strategy.
CV death was higher in the restrictive strategy.
There were no differences in adverse events.
I hope that there will be many experts who read this post. Tell me how you feel.
My take is that…
If I’ve got a post-MI patient with a hgb of 8.5, and I think blood would help, the MINT trial supports a transfusion.
But. It doesn’t compel us to transfuse. It doesn’t lend itself to algorithms or quality measures.
And that’s my favorite part. It provides evidence but allows clinical judgement.
What do you think?
I read the news late and just got to this.
My problem with this is that the study is unblinded; and the oodles of bias shows in this little stat:
>Here is another endpoint: Death due to cardiac disease was 74% higher in the restrictive arm. The CI went from 1.26 to 2.4.
This is an insane stat, and it makes me think the adjudicators of the cause of the death were biased away from blaming cardiac disease. When I declare death in patients; if I have no direct cause in mind I just put 'cardiovascular arrest'. If I were part of this unblinded study, I'd think three times before putting it for the person I just spent a couple of days filling with blood.
This then transitions over to 'adverse effects'. I am very wary of increasing blood viscosity and volume in cardiac patients; my rule of thumb is that absent defects in blood production proper, the body has the compensatory response of anemia of chronic disease for a reason.
As a non physician I don't understand why we can't create a set of X reasonable sceneries (20? 50? 200?) for each pathology/patient and based on the data we have on those, we deliver one or more "approved" therapies.
Whenever we think that there are reasons to NOT follow these guidelines we have to follow a certain process which includes an experimental framework by which the non-standard intervention and its outcomes get tracked for scientific purposes.
Isn't the risk of arbitrary decisions based on individual prejudices - worsened by the lack of precise tracking of the scenery VS intervention VS outcome - way bigger than that of being forced to deliver a bad therapy by bureaucratic-algorithmic processes?
I understand that a physician is a professional with a sophisticated knowledge base and great responsibilities, but I don't get how what you do can't be more standardized.
In this case I'd conclude that the standard of care should be transfusion, but that we need more (precise) data to understand what's going on: therefore physicians can still deliver the restrictive therapy but they have to follow a stringent experimental framework.