Article Text

Download PDFPDF
About antifragility and the challenge of dealing with endovascular therapy trials that fail to show a positive result
  1. Mayank Goyal1,2,
  2. Johanna Maria Ospel2,3
  1. 1 Diagnostic Imaging, University of Calgary, Calgary, Alberta, Canada
  2. 2 Department of Clinical Neurosciences, University of Calgary, Calgary, Alberta, Canada
  3. 3 Radiology, Universitatsspital Basel, Basel, Switzerland
  1. Correspondence to Dr Mayank Goyal, Diagnostic Imaging, University of Calgary, Calgary, AB T2N2T9, Canada; mgoyal{at}

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Endovascular therapy (EVT) is a highly effective treatment for acute ischemic stroke due to large vessel occlusion (LVO). Eight randomized controlled trials have proven its efficacy and safety in patients with LVO presenting within 24 hours.1–3 EVT eligibility criteria are likely to expand even further into the ‘fringes’, as more and more data on the safety and efficacy of EVT beyond current guideline recommendations become available—for example, for patients with M2 occlusions4 and those presenting more than 24 hours from last known well.5 In fact, it is becoming increasingly difficult to find a patient subgroup which does not benefit from EVT. While in the early days of EVT we were asking ourselves which patients to treat, the situation has now turned around; today, we ask ourselves which patients should not be treated.6

Physicians pursue a more and more aggressive treatment strategy: patient age, time from symptom onset, comorbidities, distal occlusion site, low Alberta Stroke Program Early CT Score (ASPECTS), and National Institutes of Health Stroke Scale (NIHSS) score do not discourage us from offering EVT .7 As an example, current guideline recommendations from the American Heart Association/American Stroke Association restrict level 1A treatment recommendations for EVT to patients with internal carotid artery and M1 occlusions, ASPECTS ≥6, and NIHSS score ≥6.8 However, in a recent international multidisciplinary survey with 607 physicians from 38 countries, most stated that they would offer EVT even in patients with further distal occlusions, and ASPECTS and NIHSS scores <6 (table 1).7

Table 1

Results from UNMASK EVT, an international multidisciplinary case based survey

The scenario illustrated in figure 1 is not covered by currently established guidelines. Nevertheless, we and many others think that treating this patient was the right decision. The case clearly shows a dilemma that we encounter in our clinical practice when treating those patients we consider the ‘not-so-easy-to-pluck-fruits‘ (ie, those patients in whom deficits are relatively minor, and thus the risk of making matters worse is higher).9

There is yet another dilemma, on a scientific level, that has to be solved: at present, changing treatment guidelines requires at least one, and ideally several, randomized controlled trials. Non-randomized studies, such as observational studies and registries, can also provide valuable insights and offer the advantage of larger sample sizes and include more patients, and thus probably better reflect clinical practice. Conducting them is often less expensive and tedious. However, registries yield a high risk of enrollment bias—that is, patients with complications or poor outcomes are less likely to be enrolled. Guidelines have not been changed in the past based on non-randomized trial results, and this will probably remain true for the near future.

Given the smaller effect size of EVT in patients with a low NIHSS score, low ASPECTS, etc, how do we prove the benefit of EVT in these subgroups in a randomized trial? These subgroups are also complicated by other factors, for example: a patient with an ASPECTS of 3 who is 45 years old compared with 85 years old; two patients with an NIHSS score of 2 with a P2 occlusion and hemianopsia, one of them is retired and the other is a pilot.

Our current understanding is that level 1A guideline recommendations require high level evidence from one, or several, randomized controlled trials. As a consequence, there are many ongoing stroke trials that focus on the above mentioned patient subgroups, for example the Multicenter Randomized Clinical trial of Endovascular treatment for Acute ischemic stroke in the Netherlands for Late arrivals (MR CLEAN-LATE), which aims to expand the treatment criteria for patients presenting 6–24 hours from onset (ISRCTN19922220), Efficacy and safety of ThrombEctomy iN Stroke with extended leSION and extended time window (TENSION, NCT03094715) and Large Stroke Therapy Evaluation (LASTE)for patients with an ASPECTS 0–5, and MinOr Stroke Therapy Evaluation (MOSTE) and Endovascular therapy for low NIHSS ischemic strokes (ENDOLOW) for patients with NIHSS scores of 0–5. There are initial discussions to conduct a trial focused on patients with 'medium sized' anterior circulation vessel occlusions (ie, M2, M3, large A2, and P2 occlusions). How do we react when one of these EVT trials does not show a benefit of EVT? In order to answer this question, we first have to better understand the reasons that could lead to such a neutral outcome.

Figure 1

A 58-year-old high functioning financial adviser was admitted to our emergency department. Her son called her earlier in the morning and noticed that she was not able to answer a simple question related to tax regulations, and thus called emergency medical services. At the time of arrival, the patient had no focal neurological deficits (National Institutes of Health Stroke Scale score 0), but still could not answer the tax question. CT angiography revealed a right-sided extracranial internal carotid artery occlusion with intracranial continuation (arrow in (A)). No definite changes were seen on non-contrast CT (B). Thresholded CT perfusion maps showed a large area of tissue at risk (green area in (C)) but no ’core'. The patient was taken to the angiography suite and the occlusion was swiftly and completely recanalized (D). At the moment the vessel was opened, the patient immediately spoke out and said she could remember the answer to the tax question. Follow-up MRI after 24 hours revealed small areas with diffusion weighted imaging–hyperintense signal in the right lentiform nucleus and caudate body (E, F).

Insufficient sample size

In the Highly Effective Reperfusion Using Multiple Endovascular Devices (HERMES) meta-analysis, which summarized the first major positive EVT trials, the number needed to treat to achieve at least a 1 point improvement on the modified Rankin Scale was 2.6.2 This number is almost unmatched in medicine and thus the superiority of EVT could be proven, even with relatively small sample sizes. These early trials, however, were highly selective in their enrollment criteria and focused on the ‘low hanging fruit’, as their primary purpose was to achieve a statistically significant treatment effect rather than maximizing benefit from a patient perspective. Many patients that could have benefited from EVT were excluded for the sake of maximizing the treatment effect. Now we are left with the 'not-so-low-hanging fruit'. For these, effect sizes will be much smaller, and thus the required sample sizes much larger. In other words, we need more money, better teamwork, and longer enrollment periods. The question is whether we as a scientific community are tenacious and passionate enough to meet these ambitious goals.

Complexity/lack of clarity of enrollment criteria

As the benefit of EVT has been proven in more and more patient subgroups, defining the remaining patient populations, and thus enrollment criteria of future trials, become(s) more complex and less clear. In a trial focusing on so-called non-disabling/minor strokes, for instance, it will be hard to decide whether a given patient is a suitable trial candidate because there is no good definition for non-disabling strokes. Left hand weakness might be tolerable for a right handed pensioner but will end a pianist’s career. Should we not decide on an individual case by case basis whether a stroke is disabling? Which NIHSS threshold should then be used for a trial? And should we use an NIHSS cut-off to distinguish between so-called minor and major strokes at all?

Low ASPECTS trials suffer from yet another set of problems. ASPECTS uses a dichotomous scale that is supposed to distinguish between alive and dead tissue. By common convention, a hypodense region is considered ”core”. But the decrease in density on CT scans follows a continuous spectrum rather than a dichotomous distribution, and is affected by many factors, including partial volume averaging. vasogenic edema,10 and duration of ischemia.11 Moreover, the score is subject to high interrater variability and thus not really suitable as a trial inclusion criterion.

When it comes to medium and distal vessel occlusion trials, patient outcomes and hence trial results will vary substantially between centers due to differences in operator skills. For patients in the late time window and an LVO, if they do not fit the criteria of the Diffusion Weighted Imaging (DWI) or Computerized Tomography Perfusion (CTP) Assessment With Clinical Mismatch in the Triage of Wake Up and Late Presenting Strokes Undergoing Neurointervention (DAWN)/Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke 3 (DEFUSE3) trials, there are really no other treatment options. At ours and many other institutions, we do not strictly follow the criteria in the trials and believe that they were too restrictive.

Unwillingness to randomize/lack of equipoise

Most centers around the world already treat patients who are currently enrolled in EVT trials (ie, those with low ASPECTS, low NIHSS, etc) on a regular basis,7 and thus they might not be willing to randomize patients, or only the ‘less promising’, severely comorbid patients might get randomized. Indeed, this occurred in the Acute Basilar Artery Occlusion: Endovascular Interventions vs Standard Medical Treatment (BEST) trial (NCT 02441556), which was stopped early after the enrollment of 131 patients due to an excessive number of crossovers: 22% of those patients who were randomized to the control arm received endovascular treatment. As a result, the trial failed to demonstrate the benefit of EVT over medical management in the intention to treat analysis. The per protocol analysis (ie, comparing those patients who actually received EVT against those who did not) however showed significantly better outcomes in the EVT group. It seems that many physicians did not believe that there is equipoise that would justify randomization and treated patients with EVT although they were randomized to the control arm. The most skilled operators in particular will naturally pursue the most aggressive treatment strategies and be least willing to randomize their patients, which they already routinely treat with EVT. This will most likely become a major problem in many of the currently ongoing EVT trials, especially as treatment techniques are evolving and treatment of previously inaccessible occlusions suddenly becomes safe and feasible.

Random chance

In medicine, we are excessively focused on P values and statistically significant results ( but at the same time often disregard the fact that there is a chance of not detecting a result. When the Endovascular Treatment for Small Core and Anterior Circulation Proximal Occlusion with Emphasis on Minimizing CT to Recanalization Times (ESCAPE) trial12 was designed, it was powered to achieve a type II error probability of 0.1 (ie, the probability of not detecting a significant treatment effect was 10%). In other words, there was a 1 in 10 chance that a trial would not detect a statistically significant treatment effect due to random chance, and with decreasing treatment effect sizes, future EVT trials will most likely have even higher type II error probabilities. Thus when one of the ongoing trials shows a trend towards better outcomes with EVT that happens to be not statistically significant, rather than claiming “EVT does not work in this patient subgroup”, we should think about repeating the trial and look for additional data to either confirm or refute our initial assumption.

Changing landscape may make trial results redundant even before the trial is finished

Intravenous alteplase is currently the only approved acute ischemic stroke treatment besides EVT. However, this could change soon as several new drugs have become available recently: tenecteplase is a modified variant of alteplase with greater fibrin specificity. When combined with EVT, it has been shown to have better outcomes than alteplase.13 Thus many believe that tenecteplase will soon complement, if not replace, alteplase in acute stroke treatment. Furthermore, broad implementation of EVT allows for the first time an evaluation of the neuroprotection in an ischemia–reperfusion model. The neuroprotectant NA-1 is currently under evaluation in a randomized controlled phase 3 trial (NCT02930018) and could soon complement our treatment repertoire. There is no doubt that these substances offer great hope for patients with acute ischemic stroke. However, they also add another layer of complexity to future EVT trials.

Assuming that EVT has been proven to be effective in patient population X, which was simultaneously treated with tenecteplase, can we transfer the results to population Y, which received tenecteplase and NA-1? What about patient population Z, which was treated with NA-1, but did not receive alteplase or tenecteplase? The increasing number of treatment combinations will complicate not only the design of future EVT trials but also their analysis: how much of the treatment effect can be attributed to the drug, and how much to EVT? Answering these questions will require larger sample sizes and more sophisticated statistical analyses.

EVT really does not work

This last possibility is real: maybe EVT really does not work in this particular patient subgroup. As we become more and more aggressive in our treatment approach, we will invariably reach a point at which the complication rates exceed the benefits. A genuinely negative trial result would indicate that for the given patient subgroup, this point has been overstepped. The question remains whether this result is truly generalizable to all patients that meet the inclusion criteria, or whether there are subgroups within such a neutral trial patient population that could still benefit from EVT (eg, let us assume a low NIHSS trial shows no overall benefit of EVT in patients with an NIHSS score of 0–5; there could still be benefit in the NIHSS 4–5 patient subgroup).

Assuming we face a neutral EVT trial, how can we know which of the reasons mentioned above apply to this trial, when most of us truly believe in the effectiveness of EVT in selected patients with low ASPECTS/low NIHSS/distal occlusion. Nicholas Taleb recently described the concept of antifragility: “Antifragility is beyond resilience or robustness. The resilient resists shocks and stays the same; the antifragile gets better.14 When the early EVT trials (Interventional Management of Stroke III (IMS III), etc) failed, we were collectively antifragile: rather than being paralyzed by the neutral results, we improved our methodology and could finally show he overwhelming benefit of EVT. Of note, between IMS III and the successful trials, there were dramatic improvements in workflow, and new and highly effective devices became available. Most importantly, the true effect size of EVT in the enrolled patient population was massive. This is not the case for ongoing trials. Thus we suspect that in due course, we will face a neutral EVT trial, most likely due to methodological/execution flaws (eg, an insufficient sample size and/or cherry picking). Will we then be able to maintain this degree of antifragility again? Most likely not. Thus it is imperative that the currently ongoing and future EVT trials have adequate sample sizes to detect clinically meaningful improvement in outcomes and, more importantly, make sure that experienced sites enroll all eligible patients.



  • Twitter @johanna_ospel

  • Contributors MG: conceptualization, drafting, and critical revision of the manuscript. JMO: drafting and critical revision of the manuscript and figures.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial, or not-for-profit sectors.

  • Competing interests MG reports grants from Stryker, personal fees from Stryker, personal fees from Medtronic, personal fees from Microvention, and personal fees from Mentice, outside of the submitted work. MG has a patent systems of acute stroke diagnosis licensed to GE Healthcare and a patent systems of intracranial access licensed to Microvention.

  • Patient consent for publication Not required.

  • Provenance and peer review Commissioned; externally peer reviewed.