The most recent publication of JAMA showcases two non-inferiority studies investigating the role of thrombolytics (alteplase) in acute stroke. As of today, there are two treatment modalities for an acute stroke, thrombolytics and mechanical thrombectomy. Currently, these treatments are used in combination for acute stroke.
When you arrive to the emergency room with a stroke you will be assessed for eligibility to receive thrombolytics. If you are a good candidate the ER doctor will administer it and call a specialist (like my wife 😊) to attempt a mechanical thrombectomy. In situations like this she will feed tiny instruments through a sheath inserted in a blood vessel. Feed those instruments on the end of a wire through these blood vessels to your brain. Then use those instruments to remove the clot. Hence, restoring blood flow to the brain. By the way, that sheath is about the diameter of those little red straws you use to stir your drink at the company Christmas party. Not only is the technology incredible, but the technicians are as well.
Mechanical thrombectomy in combination with thrombolytics has been shown to be superior to thrombolytics alone. However, there has been little evaluation of mechanical thrombectomy without thrombolytics. Thrombolytics are risky business. Mechanical thrombectomy is risky business. They can both result in significant harm despite optimal safety procedures. Given their potent blood thinning properties, bleeding is a major concern with thrombolytics, particularly bleeding in the brain.
This is the perfect set-up for non-inferiority. Traditionally in medical research we test things for superiority. We want to know if something is better, not equal. However, there are certain situations in which we want to prove something is just as good as the incumbent. These criteria are below.
1) The new treatment is less burdensome to patients or practitioners.
2) The new treatment is less dangerous.
3) The new treatment costs less.
When some or all of these criteria are met non-inferiority is a reasonable approach. So, in this case the addition of thrombolytics to mechanical thrombectomy is more burdensome, more dangerous, and more costly. Eliminating the need for thrombolytics would benefit patients and practitioners.
The two trials were the SKIP and DEVT trials. The trials had very similar characteristics. Both were multi-center. Skip was conducted in Japan. DEVT was completed in China. Both were randomized, open label, outcome blinded intention to treat studies.
The inclusion and exclusion criteria differed slightly. The table below shows the breakdown.
*NIH Stroke scale – stroke symptom severity scale measured (0 – 42)
^MRS- modified rankin score– functional scale (0 – 6) with 0 being completely independent and 6 is death.
+ASPECTS – (0 – 10) Score based on brain changes on CT scan with 0 being the worst and 10 being no change.
The primary endpoint was a modified rankin score (MRS) of 0-2, which would be little to no disability, at 90 days.
The DEVT trial had about 115 strokes in each arm. In the mechanical thrombectomy alone group 54% of patients had functional independence at 90 days compared to 47% in the thrombectomy + thrombolytics group. The confidence interval was [-5.1% - infinity].
There were similar death rates between the two groups at 17%. Symptomatic cerebral hemorrhage was roughly equal in the two groups at 6.1% and 6.8% respectively. Clot migration occurred in 18% of patients in the thrombectomy only arm, compared to 24% in the thrombectomy + thrombolytics arm.
None of these results were statistically significant. No superiority. No inferiority. Therefore non-inferiority. Yahtzee.
The SKIP trial showed similar results. Functional independence at 90 days was observed in 59% of the patients in the thrombectomy alone group, compared to 57% in thrombectomy + thrombolytics. However, the Odd’s ratio fell outside the range of non-inferiority with the lower boundary of 0.63 [0.63 – infinity]. The non-inferiority margin was an Odd’s ratio of 0.74. There was no statistically significant difference in death. There was no significant difference in symptomatic cerebral hemorrhage.
No superiority. Possible inferiority. Therefore, no non-inferiority. Boo double negatives!
You may be asking yourself a few questions. First, why is one end of the confidence interval infinity? Second, why was the lower bound of the confidence interval in the DEVT trial -5.1% when the reported percentages appeared superior for mechanical thrombectomy alone? Third, what is a non-inferiority margin?
The upper bound of the confidence interval is reported to infinity because there is no concern for this end of the bell curve. The investigators are not investigating superiority so a single tail statistical test is performed instead of a two-tail test.
The lower bound of the confidence interval is -5.1% even though the actual percentage appears superior due to issues of statistical power. The confidence interval shows the probability that the true result lies within it based on relative event rates in the groups. When it is wide like this, it is an indication that the true result is fragile.
The non-inferiority margin is the expected reduction of efficacy that would be considered clinically irrelevant. The DEVT trial used a non-inferiority margin of 10% absolute risk increase. They were willing to accept 10% of individuals losing functional independence due to a stroke to avoid the need for thrombolytics. They didn’t really explain the origin of the 10% number. It read like a wet thumb in the air judgement to measure the direction of the wind.
This is a wide margin for a disabling disease like stroke. I would not consider that non-inferior. This would be a number needed to harm of 10. Usually, we’re jumping up and down when we find something that helps 1 in 10 patients, and apologizing vociferously when we harm 1 in 10. The risk of symptomatic cerebral hemorrhage from thrombolytics is roughly 5%. That’s a number needed to harm of 20. At minimum the clinically acceptable difference should be 5%, not 10%.
The SKIP trial also failed to meet its margin. However, this is not due to it being broad. They were willing to tolerate at 26% increase odds that the patient would do poorly. The lower end of the confidence interval included a 37% increase in odds. The failure in this study was mainly due to statistical power. The result had a wide confidence interval likely due to the small study population and narrow inclusion criteria.
SKIP based their margin off of previously conducted research. Had SKIP selected to use the 10% absolute risk margin they likely would have achieved non-inferiority. However, the result would have been scrutinized as DEVT for having a broad interpretation of minimal clinical relevance.
It’s difficult to know at this time if these studies will impact clinical practice. Thrombolytics is a prime candidate to be scrutinized. In narrow populations like this the exact benefit is not known. They increase risk, burden and cost without definitive benefit. But as you can see, non-inferiority is nuanced. The non-inferiority margin is rarely established by consensus. The researchers are left with a “thumb in the wind” guestimate to what will be accepted by their peers.
Time will tell if thrombolytics remain part of the stroke treatment pathway. My guess is for now they will continue as one of the cornerstones for stroke treatment, especially in cases that would not have met criteria to be in these studies.
Now you should be able to approach a non-inferiority trial with confidence. The proposed intervention should reduce burden, improve safety, and reduce costs. The non-inferiority margin should truly be clinically irrelevant or established by prior research. Once these items are considered it’s just like any other trial.