Discovery Channel had Shark Week. Around Emergency Literature of Note Headquarters, we do Stroke Week.
Why? Because, from a methodologic standpoint, acute stroke care is the gift that keeps on giving. This week, we will see randomized-controlled trials stopped for “loss of equipoise”, a few authors who are still working out how to conduct a systematic review, and more practice-changing conclusions drawn from retrospective registry data.
As we noted a couple weeks ago, there were three major endovascular trials presented in early February – ESCAPE, EXTEND-IA, and SWIFT-PRIME. ESCAPE and EXTEND-IA were simultaneously published in the New England Journal of Medicine. Now, SWIFT-PRIME has reached final publication. These trials are hailed as a sort of second-coming of the messiah for the exiles wandering in the wilderness since PROACT-II.
And, now, interestingly, a fourth trial is presented – simultaneously published along with presentation at the European Stroke conference. This is REVASCAT, yet another stent retriever trial funded by an unrestricted grant from Covidien, the manufacturers of the Solitaire device. And, I can tell you Covidien saved themselves a lot of money in this trial – because it was planned to enroll 692 patients, and was terminated after 206.
Why was it terminated? Not, as the other trials were, due to having met pre-specified efficacy criteria. This trial was stopped because of “loss of equipoise”, following presentation of the other trials. This is, effectively, the equivalent of stopping your moon landing program because the other side got there first, sitting around glumly shuffling papers. But, more data is still important – and this data is important because it throws a little bit of cold water on the other trials.
SWIFT-PRIME, for example – mRS 0-2 in 60% of the endovascular intervention cohort, compared with 35% of the tPA-only cohort. REVASCAT – mRS 0-2 in 44% of the endovascular cohort, compared with 28% of the tPA-only cohort. 25% treatment difference versus 16% treatment difference. SWIFT-PRIME – 12% mortality in the endovascular cohort, compared with 26% mortality with usual care. REVASCAT – 18% mortality in the endovascular cohort compared with 15%.
What’s different? Where the previously presented trials used strict imaging criteria for small infarct cores and good collateral circulation, REVASCAT simply included all patients with low ASPECTS scores and proximal vascular occlusions. This is, then, more akin to MR-CLEAN or ICARO-3, in which the benefit is attenuated substantially if the status of the underlying tissue is not fully appreciated.
The lesson from this should be clear – imaging criteria requiring salvageable tissue as result of collateral flow provide maximum yield in reducing the number of endovascular procedures performed with low or no chance of benefit. Whether these lessons are heeded, I remain highly skeptical.
The other lesson: when you’re hot, you’re hot, and even lukewarm half-raw results can still get you into NEJM.
“Thrombectomy within 8 Hours after Symptom Onset in Ischemic Stroke”
http://www.nejm.org/doi/full/10.1056/NEJMoa1503780
“Stent-Retriever Thrombectomy after Intravenous t-PA vs. t-PA Alone in Stroke”
http://www.nejm.org/doi/full/10.1056/NEJMoa1415061
I am unsure of your thesis. Is it the case that you believe that additional observations are required to state confidently that endovascular treatment for acute stroke is beneficial under the conditions examined in the various trials to date? If you believe this is the case (as did these trialists) then I do not think that further randomization of eligible patients to 'best medical therapy' is ethically justifiable. If you do not believe this is the case, then might I ask why? To date we have accumulated five positive trials, all with highly statistically significant results. Although premature stopping increases the probability of an inflated effect size, I suspect that it is very improbable that this bias is driving the treatment effect (the math is a bit beyond me so hopefully a statistician can comment).
There is of course work to be done to better define the edges of the patient population who benefit from this treatment, but it would be statistically inefficient (and I believe unethical) to try to do that with the existing trial designs.
It depends on your threshold for ethical randomization. Is the goal to avoid harm to a handful of patients enrolled in a clinical trial? Then, yes, it is unethical to continue to randomize those patients once the pre-determined threshold for benefit has been achieved on a pre-planned interim analysis. Is your goal to inform the safest, highest-yield implementation of a medical therapy for hundreds of thousands of stroke patients for decades to come? Then, in the interests of public health and well-being, it is ethical to continue randomization.
There is a great difference between “showing benefit” and understanding how to appropriately target a therapy. It is, perhaps, reasonable stop a trial such as REVASCAT when it becomes obvious the intervention is inferior to a better-designed intervention, say, the trials with collateral circulation. However, if the editorial in the NEJM is any indication of the enthusiasm for endovascular therapy, I expect the upcoming guidelines to include a vast anterior stroke population unselected by perfusion imaging – a lowest common denominator – rather than a well-designed recommendation balancing benefit and resource utilization.
Hearkening back to those bioethics days, my fuzzy recollection of the concept of clinical equipoise is that the treating clinician should have uncertainty regarding the expected utility of two different treatments for the particular patient that they are faced with. If I believe that enrolling a patient in a clinical trial will more likely cause harm than benefit to that patient, then I should not enrol them (regardless of any potential value to society). As a concrete illustration that I posted on Rory Spiegel's page:
As an example – continuing the ESCAPE trial to planned enrolment of 500 participants would have resulted in approximately 92 participants being exposed to best medical therapy when they might have otherwise received endovascular treatment (as they would now). Based on our best available estimates from all five trials (NNT range 4-7), this would have resulted in avoidable disability in between 13-17 patients. I do not myself think this is justifiable.
As a thought experiment, I considered what I would say to a patient facing enrolment in such a trial today, and what I would say if I were the patient in that situation. As a practitioner, I would feel a duty for full disclosure. As a (trial-eligible) patient, I cannot imagine declining endovascular therapy given all that has been observed to date. I believe that continuing this trial flunks the ethical "smell test."
Pre-determined thresholds for pre-planned analyses are helpful for reducing the risk of bias – and I do acknowledge that there is some risk of bias in our estimates of the treatment effect. However, I disagree that the magnitude of this bias is relevant given the large treatment effect size generated from many observations across multiple independent trials. I think that the slight increase in bias from a premature analysis (to which I imagine the authors were committed – I doubt this was a situation of "peek-and-if-negative-continue") is justified given the ethical problems that I see in continuing.
We will all have an opportunity to evaluate guideline statements once they are published, and we can be hopeful that they will keep a lid on indication creep. Of note, the ESCAPE trial did not require perfusion imaging – it relied on a scan-rescan technology that can be implemented using non-volumetric scanners.
To some extent, I think you've reinforced the best argument in favor of continuing these trials – that continuing some of these trials would have resulted in 13-17 patients with potentially avoidable disability. And, with your off-the-cuff NNT calculations, we then arrive at 3-6 patients subjected to the costs and harms of endovascular intervention without a measurable benefit in order to improve one patient outcome.
These are still pretty good NNTs! But, if you multiply a therapy poorly informed by reduced sample sizes in the initial wave of trials by decades of therapy, you end up harming far greater numbers of patients than the 13 to 17 you find it unethical to expose.
Think of tPA in acute ischemic stroke. If NINDS were 5,000 patients instead of 500 patients and showed the same magnitude of benefit, the skepticism regarding this treatment would be substantially reduced – and far more patients would have benefited from this treatment over the last two decades. The narrow view here turns out to be the more harmful view in the long run.