JC: Kicking against the prick: Systematic Review of stroke thrombolysis

I think it’s probably fair to say that stroke physicians and emergency physicians disagree about the utility of thrombolysis in acute ischaemic stroke. They love it, probably because of its occasionally amazing effects in restoring an apparently dense hemiplegia with aphasia to almost normality. We hate it, because we see the effects of symptomatic intracranial haemorrhage and the early deaths as a result.

Its also probably fair to say that we disagree over our interpretation of the evidence. Despite thrombolysis being used in many forms and in small trials since the 1960s, it didn’t really gain prominence until the publication of the NINDS trial in 1995. This has been discussed elsewhere but to recap, in brief, it was a trial of human recombinant tissue plasminogen activator (tPA) versus placebo in presumed acute ischaemic stroke, given within 180 minutes of symptom onset. This showed improved outcomes on various functional scales, but a significantly increased rate of symptomatic intracranial haemorrhage and early mortality in those treated with tPA.  Greater minds than mine have questioned not only the methodology and results of the NINDS trial (Hoffmann 2001, Mann 2002) but also the implementation of the trial in the real world (Katzan, 2000). There were also significant concerns over the involvement of the drug manufacturer in the process of authoring the manuscript.

The 2009 Cochrane review on stroke thrombolysis (written by a stroke physician in an eminent neurosciences centre) concludes that thrombolysis “is one of the most promising treatments for acute ischaemic stroke” but that it raises more questions than it answers:

“How big is the overall benefit? What is the latest time window in which the treatment is still beneficial? Which grades of stroke severity and which types of stroke, as judged clinically and on brain imaging, are more likely to respond favourably to treatment? Should patients aged over 80 years receive thrombolysis? Which types of patients are most likely to be harmed by, and which to benefit from, treatment (e.g. with or without other major medical conditions like cardiac arrhythmias, diabetes, hypertension, or other disorders and concomitant medication)?” (Wardlaw, 2009).

This was before the publication of the IST-3 trial in 2012, which was the biggest trial of tPA versus placebo in presumed acute ischaemic stroke. IST-3 was significant, not only for the numbers (n=3035), but also because it included 1617 patients over 80 and extended the tPA window to 6 hours, both of which better reflect real world medicine in both population and pragmatism.

IST-3 again was polarising between stroke physicians and emergency physicians (Hoffmann 2012, Radecki 2012)12. There was no significant increase in those alive and independent at 6 months (554 (37%) patients in the rt-PA group versus 534 (35%) in the control group), but there were 7 times more fatal or non-fatal SICH in the treatment group (104 (7%) vs 16 (1%)) and a 50% increase in 7 day mortality (163 (11%) vs (107 (7%)). The abstract rather cheekily suggests that “between 7 days and 6 months there were fewer deaths in the rt-PA group than in the control group, so that by 6 months, similar numbers, in total, had died (408 [27%] in the rt-PA group vs 407 [27%] in the control group)” and concludes “thrombolysis within 6 h improved functional outcome. Benefit did not seem to be diminished in elderly patients.” Convinced? No, me neither.

Accordingly, it was with great delight that this paper appeared in all of our inboxes to try and put to bed the debate over thrombolysis in stroke for good.

This is set out as an independent (presumably free from commercial tie and not involved in the fracas between stroke physicians and emergency physicians) systematic review and meta-analysis to assess the harms and benefits of intravenous thrombolysis for patients with presumed acute stroke. It’s also a mammoth piece of work. The paper itself is 15 pages but the supplementary material is 41 MB of goodness and it’s all #FOAMed.

So what sort of study is this?pyramid of evidence

This is a systematic review and meta-analysis of all the papers published about thrombolysis in stroke. It’s really two papers in one – a systematic review can stand alone, but you cannot have a meta-analysis without the systematic review. In brief, a systematic review is an attempt to summarise, criticise, synthesise and interpret the results. More on this here. A meta-analysis attempts to go one step further to provide a quantitative estimate of the net benefit of the treatment when all the reviewed studies are aggregated. More on this here. As this is both a very powerful tool and the apotheosis of EBM, there are rules around it’s use. Firstly, it should be registered and preferably have the protocol available for all to see. Second, it should be presented like a “normal” paper with abstract, introduction, methods, results, discussion. Third, there should be a search strategy that anyone can replicate to check. Fourth, there should be a described rationale (a priori) to select papers that then should be rated and assessed for bias. Data should be extracted, preferably to a (web) published form and then synthesised. These should be then presented and discussed. The PRISMA website is your go to resource for all of this. Sounds straightforward? It bloody well isn’t. I have done several, including some published ones and they are hard work. Just a comprehensive search strategy alone is tricky and you always find that you have missed a paper unless you are really careful. This always comes back to bite you.

Anyway, tProspero and miranda.jpghis is a suprememly well done SR and MA. Why? Well, for one it was registered . Everything that the team wanted to do beforehand was planned and registered. No post hoc theories or dubious data trawls here. There was also an amazingly thorough search strategy, including such agents as lumbrokinase and saruplase (nope, me neither). The studies were selected for inclusion by previously published criteria and assessed for risk of bias using an approved tool. Given that many of the papers about stroke thrombolysis were heavily (and at times surreptitiously) sponsored by Big Pharma, this was also recorded. Then the handles were cranked and the elves got to work (which is a shorthand way of describing monumental and painstaking effort of entering lots of data into a meta-analysis engine such as the freely available RevMan from Cochrane).

So what did they look at?

The primary outcome was good functional outcome at final follow up. This has multiple definitions but the chief one was a modified Rankin score of 3 or less – able to walk and care for personal needs but may require some assistance, which is probably not a bad place to be after a dense hemiplegic stroke, but less good if you merely had transient dysarthria.

So what did they find?

From an initial 20,296 studies they whittled these down to 26 with 10,431 participants. These are described in detail in table 2 of the paper. The authors report important caveats that I quote verbatim:

of the 16 studies that nominated a specific primary outcome, only two studies reported a significant treatment effect in favour of thrombolysis and five studies were stopped early after interim analyses demonstrated either harm or lack of benefit… The largest study did not use any method of blinding, and ascertainment of the primary outcome was conducted in an unblinded fashion via telephone or postal survey. Both studies reporting a positive treatment effect based on the primary outcome reported significant baseline imbalance. While the majority of studies were sponsored by pharmaceutical companies, the most influential study did not declare any such sponsorship. Notably, few studies provided details regarding anticoagulant and antiplatelet agent use after 24 h, nor did they specify that patients were cared for in a stroke unit post randomisation. Only four studies had specific protocols for the management of blood pressure

Now, I am an admitted cynic about thrombolysis in stroke, but surely even the most ardent believer must have at least a fleeting doubt when the treatment is summarised like that?

So what did they find?

What was likely a huge amount of work is summarised in table 3 in the manuscript.


This table allows us to do several things. The first is to state categorically that all thrombolytics (and tPA is no exception) cause a significantly greater risk of symptomatic intracranial haemorrhage and early mortality. The second is to state that all thrombolytics and tPA lead to a better functional outcome at final follow up (possibly because of increased early mortality). The third is that there is an increase in mortality at final follow up for all thrombolytics and a trend to increased mortality at final follow up with tPA. That tPA is slightly “safer” than all thrombolytics will surprise none of us who had the pant-filling terror of the weird dysrhythmias joys of using streptokinase. We can also work out the NNT for tPA for a good functional outcome. The figures go as follows. For tPA versus placebo the rates of good functional outcome are 1919/3129 (61.3%) and 1728/3045 (56.7%). This give an Absolute Risk Reduction of (61.3 – 56.7) 4.6%. NNT is 100/4.6 = 21.7

I don’t feel that the table does the authors justice. Sure, the paper is long (and EPs have a short attention span and there is a danger of tl; dr) but the true genius (and the reflection of the graft put in by the four authors) is unfortunately hidden away in the supplementary material and it is for this the authors should be praised as it is truly fascinating.

How about this?

For tPA vs control, the rates of symptomatic intracranial haemorrhage are 273/3674 (7.43%) and 62/3513 (1.76%). That gives an Absolute Risk Increase (ARI) of (7.43 – 1.76) 5.67%. NNH (number needed to harm) is therefore 1/5.67% = 17.6

For early mortality, the rates are 250/2807 (8.97%) and 174/2728 (6.38%) respectively. ARI = 2.54%. NNKill = 39.5

Have a think on that for a minute. For every 18 patients treated with tPA over placebo, 1 will get a SICH and we all know that there is likely minimal treatment available. For every 40 patients treated with tPA one extra will die. Primum non nocere? It’s no surprise many bodies are moving away from stroke thrombolysis. Treat 200 patients, 10 extra will get a good functional outcome, but 5 extra will die in the first 10 days. There may not be any effect on 185 patients. Alteplase costs between £300 to £600 to deliver, not counting staff and disposables. That’s around 10 grand per good functional outcome and 18 grand to kill someone. We’ve spent £83250 for no discernible effect in the other 185.

It gets better in the long term. For mortality at final follow up the NNKill is 333. However in the volume of patients treated with thrombolysis for stroke internationally, that is a significant number of excess deaths. In the UK there are in excess of 152,000 strokes annually. If all of these are thrombolysed (unlikely I know) then that’s over 450 excess deaths.

So what’s the bottom line?

The authors conclude that

The current summary of evidence for the use of thrombolysis in presumed acute ischaemic stroke shows clear early harm, in terms of increased rates of symptomatic intracranial haemorrhage and increased early mortality. There is no evidence of late reductions in mortality, but an improvement in late functional outcomes that is largely reliant on one small trial and one large trial with significant methodologic limitations. As such, it is very likely that those skeptical of the relative benefits of this therapy have foremost in their minds the avoidance of adding further harm to patients with an already devastating condition. It should be noted although that withholding a treatment that could lead to improved functional outcomes might also leave patients with significant functional deficits that could have been avoided. The available data are unlikely to resolve this controversy, and replication studies with robust methods are urgently required.

I’m sceptical, of that I make no secret. However, I would sum up my concerns like this:

  1. Every medical student learns that one of the cause of confusion is an acute stroke. Indeed, one of the proponents of stroke lysis tried to popularise the term “brain attack” to elevate to parity with heart attack. If your patient is has had a stroke, prior to administering the NIHSS (which I have NEVER seen done by a stroke physician), do you assess for confusion? If so using what instrument? How do you communicate the risk, given that doctors are bad at both estimating and conveying it, and there are still several million who don’t understand risk and buy a lottery ticket? Do you mention the NNT, NNH and NNKill mentioned above? Do you even know them? How then is the consent that you have taken valid? This isn’t an accusation, it’s a question.
  2. Say a member of my family has a  hemispheric stroke and are suitable for thrombolysis:
    1. would I recommend tPA for my 85 year old granny with IHD and arthritis? Hell no
    2. would I recommend it for my 62 year old father who’s well, retired and active? No, but it’s his choice.
    3. would I have it myself? I don’t really know. Now when I’m well, definitely not, but if I was in the ED, with a time window and dense hemiplegia on my dominant side, then even knowing the harms and minimal benefit, then maybe, just maybe. And I’m quite bright and I’ve read the evidence. So how do we expect a lay person (such as my father) to make this choice?

And this for me is the crux. It’s not a question of science. The science suggests that thrombolysis for stroke should be sent back to the clinical trial stage. There is no evidence in this, the best review and meta-analysis to date of the subject, to recommend it. The emotive issues around stroke persist, however. I completely understand the desire of the stroke physician to “not just stand there, but do something”, especially in the presence of a young stroke victim, much in the same way that I feel the need to do something to the child pedestrian hit by a car. However, unlike my magic bullet of TXA, which is cheap and harmless, theirs is worthless and expensive. The real need for us both however, is to organise and agitate, her for organised, resourced stroke centres and further primary prevention and mine for speed limit restrictions, road furniture and crossings.



Before you go please don’t forget to…


Hoffman J, Cooper R. How is more negative evidence being used to support claims of benefit: The curious case of the third international stroke trial (IST-3). Emerg Med Australas. 2012;24(5):473-476. [PubMed]
Radecki R, Chathampally Y, Press G. Annals of Emergency Medicine Journal Club. rt-PA and Stroke: does IST-3 make it all clear or muddy the waters? Ann Emerg Med. 2012;60(5):666-667. [PubMed]


  1. Pingback: JC: Kicking against the prick: Systematic Review of stroke thrombolysis – Global Intensive Care

  2. Derek Louey

    Even with robust data such as this which calculates the average statistical outcomes for all-comers, the difficulty of providing informed consent is magnified by attempting to calculate the risk to the individual.

    We know there are a myriad of other factors that affect the risk:benefit balance. They could lower the chances of success, increase the chance of complications or more importantly, predict spontaneous recovery without any intervention required.

    I have rarely seen a much more nuanced discussion as this occurring. More often than not I just hear a PGY 3 medical resident give an over-simplified explanation of ‘it could make things worse but this is the best chance he has of getting better’.

  3. Roberto Cosentini

    Hi Simon
    Great post!
    I think we should focus on two goals.
    One is to identify the subgroup of patients who would benefit most by thrombolyses (e.g.: younger and non-diabetic subjects) by individual patient meta-analysis, for instance.
    The second is a shared decision with patients and or family putting these figures in the discussion (NNT 21, NNH 18, NNK early 40, NNK 333). I would also suggest to add the local figures according to volume and experience of the site of treatment.
    Thanks again for the contribution to knowledge reflection and spreading.

  4. Pingback: Ischemic Stroke Treatment Archive - R.E.B.E.L. EM - Emergency Medicine Blog

  5. Jo

    Thanks for that Alan. I’d read the abstract before but it didn’t have the numbers below the headlines. I have to admit it makes consenting people for thrombolysis (done by ED doctors at our hospital) much harder. Not just because I didn’t have the numbers before but also because because my heart isn’t really in it. Had a rather surreal experience the first time I consented after reading the abstract: “we have this treatment that might make you better, it is more likely to kill you in the next week, but if it doesn’t you are less likely to be disabled in the long term” I thought I phrased it as “don’t have it you’ll die” the family heard “it will make you better and the doctor wouldn’t be suggesting it he didn’t think it was a good idea” and they consented!
    The problem, as alluded to by an earlier reply similar to all shared decision making is that trials give us population risks but not local risks and certainly not risks for the patient in front of you. “If we thrombolysed 200 people….” they don’t care about 200 people, they want their risk based on their age, comorbidities, onset time. We don’t have that. It’s what appeals about the MACS rule.

  6. Joe

    I had to check the date on the top of this to make sure it’d been written relatively recently. It has, and so I’m very surprised you haven’t mentioned thrombectomy? I’m a bit of a tPA skeptic as well, but if I or anyone I know had a proper ischaemic stroke (and by ‘proper’ I mean NIHSS > than about 6, and a clot in a vessel on a CTA) I’d be getting them to the nearest neuroIR centre as quickly as possible. The HERMES meta-analysis published in the Lancet last year of the big 5 thrombectomy trials showed that the NNT to reduce disability on the Rankin scale by one level in these patients was 2.6!

  7. Derek Louey

    But what proportion of all-comers are eligible for the intervention and what implications would this have on the health network if all stroke patients were only delivered to neuro-intervention centres



Thanks so much for following. Viva la #FOAMed

Translate »