Last year, the FDA granted andexanet alfa an accelerated approval as a reversal agent for factor Xa inhibitors, but based that decision on a complete lack of useable science. This decision is almost identical to the approval of idarucizumab, a debacle I have discussed previously. (At that time, I was chastised by Scott Weingart for saying “this is not science”. Because the activity of uncontrolled observation certainly fits into a broader definition of science, I will refine my terminology and say instead that this is not usable science). This paper is another bit of advertising disguised as (completely unusable) science in the increasingly horrible New England Journal of Medicine.
Connolly SJ, Crowther M, Eikelboom JW, et al. Full Study Report of Andexanet Alfa for Bleeding Associated with Factor Xa Inhibitors. N Engl J Med. 2019 [free full text]
This is a multicenter, prospective, open-label, single-group study.
Adult patients with major bleeding who had taken one of apixaban, rivaroxaban, or edoxaban at any dose or enoxaparin at a dose of at least 1 mg per kilogram per day within the last 18 hours.
- Major bleeding was defined as signs of hemodynamic compromise, or a hemoglobin drop of 2 g/dL (20g/L), or any hemoglobin blow 8 g/dL (80 g/L), or bleeding in a critical area (retroperitoneal, intraarticular, pericardial, epidural, or intracranial bleeding or intramuscular bleeding with compartment syndrome).
- Exclusions: Surgery planned within 12 hours, intracranial hemorrhage with a GCS less than 7 or an estimated hematoma size of 60 mL, expected survival less than 1 month, a thrombotic event in the last 2 weeks, or any of the following in the prior 7 days: vitamin K antagonist, dabigatran, prothrombin complex concentrate, recombinant factor VIIa, whole blood, or plasma.
Bolus of andexanet alfa followed by a 2 hour infusion.
2 primary outcomes:
- The percent change from baseline in anti–factor Xa activity. (This was originally a secondary outcome, but they changed it to a primary outcome).
- Percentage of patients with excellent or good hemostatic efficacy 12 hours after the andexanet infusion.
They enrolled 352 patients. There were 128 patients (36%) receiving rivaroxaban, 194 (55%) receiving apixaban, 10 (3%) receiving edoxaban, and 20 (6%) receiving enoxaparin. Of the 352, only 254 were included in their efficacy analysis, because 100 patients had low anti-Xa blood levels. (Even in a carefully controlled clinical trial, with access to anti-Xa levels, 28% of the population treated had no indication for the drug. It makes one wonder how this would play out in the real world. My guess is andexanet alfa will be given to a lot of people it couldn’t possibly help).
Anti-factor Xa activity was reduced in all groups. The numbers probably aren’t important, because we have no idea if this lab measure is clinically important, but if it is a valid surrogate, the numbers all decrease a lot initially (although weren’t as impressive at 4-12 hours).
82% of the cohort were judged to have “good or excellent hemostasis”, based primarily on size of hematoma increase or hemoglobin drop.
34 patients (10%) had a thrombotic event in the 30 day follow up.
There were also 49 deaths (14%) at 30 days, despite patients with expected survival of less than 1 month being excluded from the trial.
Although this data is being used to support the FDA approval, and will unfortunately will lead to this experimental chemical being used on unsuspecting patients, there is no useable science here. (I really could just cut and paste the discussion from my post on idarucizumab, because the fundamental flaws are the same.)
Anti-factor Xa activity is a surrogate outcome with unclear implications. We don’t have strong data supporting a correlation between anti-Xa activity and patient outcomes, and even if we did, that wouldn’t guarantee that lowering anti-Xa activity pharmacologically would improve outcomes. (We know high INRs are associated with bad outcomes in trauma, but outcomes remain pretty bad even when we correct the INR. Likewise, low pH is bad in DKA, but trying to fix the number with bicarb doesn’t seem to help patients.) Just because a surrogate outcome is associated with clinical outcomes doesn’t meant that artificially altering that surrogate number will help patients.
To make matters worse, they actually looked at the correlation between anti-Xa activity and hemostasis and found none.
“Overall, there was no significant relationship between hemostatic efficacy and a reduction in anti–factor Xa activity during andexanet treatment.”
Of course, they do report clinical outcomes here, but the numbers are meaningless. They specifically chose a relatively healthy cohort (by excluding anyone who might die), and then declare the outcomes were good. (Keeping in mind that the trial was not blinded and was designed by the drug company).
However, without a control group, it is impossible to know what the numbers mean. Although 82% of this population had “good hemostasis” when given the experimental agent, it is quite possible that 90% would have had good hemostasis without the drug. There is no way to know.
Even worse, in a population that was specifically chosen to exclude patients who might die, 14% of the population died. We have no idea if these patients would have died anyway, but there is clearly a possibility that andexanet alfa is killing patients. Similarly, 10% of this population had thrombotic events.
In the end, we have no idea if this agent helps patients. We know it is expensive ($20,000 to $55,000 a dose). We know that if given, we will see some biochemical changes. We know a lot of patients will stop bleeding, but have no idea if they would have stopped bleeding anyway. We know there will be thrombotic events. We know that if it is given in a population of patients you don’t expect to die, 14% will die.
I think it is very clear that we should not be using this experimental chemical on our patients. Most medical practices are not parachutes. Andexanet alfa is clearly not a parachute, as 14% of a healthy population still died. Human physiology is complex, which makes it very difficult to predict the outcomes of our interventions. As a rule, all of our actions can be harmful. The trick is to ensure that benefit is likely to outweigh that harm. That is why we need science. And in the case of andexanet alfa, we do not have any usable science.
As an aside, when designing this trial, they decided that exposing people to placebo would be unethical. The problem with that argument is that it assumes the therapy works, which is generally a poor assumption in medicine. In fact, if anything is unethical, it is allowing studies like this to be conducted. All trials put participants at risk. Study participants volunteer to accept the risk of an untested, experimental chemical, with the hope that society will benefit from the resultant science. This trial put patients at risk without any hope of usable clinical data. (If all you wanted was the biochemical data, a much smaller phase 2 trial would have sufficed.)
I think this point is highlighted by the fact that, despite this supposedly positive trial, the FDA has demanded an RCT, which is supposed to start soon. This is a surprisingly appropriate response from the FDA, but it does contrast strangely with the full approval given to idarucizumab based on similarly garbage observational data. I’m not sure how to explain the difference.
Andexanet alfa should not be used clinically based on this data. We should wait until we see the results of RCTs.
Some Twitter Discussion
I didn’t harp too long on the financial conflicts of interest in this paper, but they are significant:
I have written at length about my approach to critically appraising medical evidence. However, I now have a new approach when it comes to papers found in the New England Journal of Medicine:
Other Andexanet Alfa FOAMed
Also, for a discussion of some of the other articles in this issue of the NEJM, and a further explanation of why I came down so hard on this journal, see “The Case of the Scientific Ruse” on EMNerd.
Morgenstern, J. Andexanet Alfa: More garbage science in the New England Journal of Medicine, First10EM, February 11, 2019. Available at: