Overview of Cochrane thrombolysis meta-analysis

Objectives.

We wished to determine whether (and in what circumstances) thrombolytic therapy might be an effective and safe treatment for acute ischemic stroke, and in particular to determine (a) the risk for death within the first 2 weeks and at long-term follow-up, (b) the risk for early symptomatic or fatal intracranial hemorrhage; and (c) the proportion of patients with a poor functional outcome (i.e., dead or dependent). We also wished to undertake exploratory analyses to examine the effect of antithrombotic therapy given concurrently, the severity of the stroke, and the time window to treatment on the balance of risk and benefit with thrombolysis.

Methods.

We sought to identify all truly randomized trials of thrombolytic therapy compared with control in patients with acute ischemic stroke.

The search strategy used methods developed for the Cochrane Stroke Group,5 as well as searching Embase, hand-searching neurologic and stroke journals in English and non-English languages (particularly in Japanese and Chinese), writing to 321 pharmaceutical companies for published and unpublished information, and searching meeting abstracts and attending relevant conferences.

Eligible trials were those that included patients of any age or sex with a definite acute ischemic stroke (CT scanning having excluded cerebral hemorrhage), or those involving testing of any thrombolytic drug. Trials in which the treatment or control group received another active therapy not factored in to the randomization (e.g., thrombolytic drug plus another agent versus placebo, or thrombolytic drug versus another agent) were excluded. Trial quality was assessed on (a) the method of randomization, (b) blinding of treatment administration, and (c) blinding of outcome assessment.

Data were extracted from each included trial based on the number of patients originally allocated to each treatment group for an intention-to-treat analysis on the prespecified outcomes: (a) deaths from all causes within the first 7 to 10 days after treatment, (b) symptomatic (i.e., associated with a deterioration in the patient’s neurologic state), or fatal (i.e., leading directly to death) intracranial hemorrhage, (c) total deaths (including early deaths) from all causes at long-term follow-up; and (d) “poor functional outcome” at the end of follow-up.

When data were not available in the publication, additional information was sought from each trial principal investigator. “Symptomatic intracranial hemorrhage” included bleeding into the infarct or new bleeding at an anatomically separate site elsewhere in the brain or its surrounding spaces after randomization, confirmed by CT scanning or postmortem. “Poor functional outcome” was defined as death or dependency, measured by the Rankin or Barthel scale, at the end of trial follow-up. “Dependency” in the present analysis was defined as a score of 3 to 5 inclusive on the Modified Rankin Scale. Some would prefer a definition of “good outcome” (independence) including Rankin 0 and 1 only. Therefore, wherever possible, data were sought on the number of patients in each individual Rankin category to allow an analysis by either definition of functional outcome.

Data were also extracted based on the proportion of patients given aspirin (ASA) or heparin within the treatment period, and the time window to treatment (within 3 hours of the stroke or later than 3 hours). A tabulation of the extracted data was cross-checked and then verified with the principal investigator of each trial. The analysis was by odds ratio (OR) and 95% confidence intervals (CI) and by absolute effects per 1,000 patients treated.

Results.

Methodological quality of the included trials. Seventeen trials (5,216 patients) are included. Only the key differences are highlighted in the text, full details being given in the table⇓ and the CDSR.5 It should be noted that the National Institute of Neurological Disorders and Stroke (NINDS) trial6 was conducted in two consecutive parts, A and B, but published in one paper. As a result, it is included as one trial in this review. The results of Alteplase Thrombolysis for Acute Noninterventional Therapy in Ischemic Stroke (ATLANTIS) A and B and Recombinant Prourokinase in Acute Cerebral Thromboembolism (PROACT) II7,8⇓ were obtained from oral presentations and therefore must be regarded as preliminary. Data on about 72 patients in ATLANTIS part B were not included in the presentations. Therefore the total number of patients included in this review is 5,144. Four trials used i.v. streptokinase (SK),12,17,18,20⇓⇓⇓ eight used i.v. recombinant tissue plasminogen activator (rt-PA),6,7,13-16,19⇓⇓⇓⇓⇓⇓ three used i.v. urokinase (UK),9-11⇓⇓ and two used intra-arterial recombinant pro-urokinase (r-ProUK).7,21⇓ Thus trials using i.v. rt-PA contribute 2,889/5,144 patients, or 56% of the data.

The trials performed in the 1980s9-11⇓⇓ were methodologically very different from all the trials performed later (table⇑). However, the 1980s trials9-11⇓⇓ did not collect data on functional outcome and therefore only the 1990s trials6-8,12-21⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓ contribute to the analysis of death or dependency.

Most trials measured the severity of the stroke at baseline. All trials excluded unconscious patients and most trials did not randomize many drowsy patients, except MAST-E,17 in which 50% of the patients were drowsy or stuporous at randomization. Only five trials had no upper age limit and included very elderly patients9-11,17,18⇓⇓⇓⇓ (patients over the age of 80 were randomized in the NINDS trial6 but the actual age of the oldest patient was not stated in the primary publication, although a patient of 87 years is mentioned in a subsidiary paper). Several trials excluded patients with another aspect of visible infarction on their pre-randomization CT,7,8,13,14,16,19,21⇓⇓⇓⇓⇓⇓ but others did not. However, individual doctors may have excluded patients with visible infarction, depending on local opinion.

The randomization method varied: five trials used central telephone randomization,8,12,17,18,21⇓⇓⇓⇓; three used selection of a sealed, sequentially numbered prepack (of active drug or identical-appearing placebo) at the participating hospital, followed within 2 hours by a telephone notification to the Central Trial Co-ordinating Office6,7⇓; seven used sealed drug prepacks of active drug or identical-appearing placebo numbered from a randomization schedule drawn up centrally9-11,13,14,16,19⇓⇓⇓⇓⇓⇓; one used sealed envelopes15 and the method was not stated in one other.20

It was found that antithrombotic drug use varied. Only one trial factored ASA use into the randomization and therefore was the only trial thus far to test for an interaction between thrombolytic and antithrombotic drugs.18

The timing of final outcome assessment varied between 1 and 6 months after the stroke (table⇑). It should be noted that follow-ups at 6 months and 1 year have recently been reported for the NINDS trial6 but that the 3 month (primary) outcome is used in the present review. Only four trials used blinded, independent follow-up (one by central telephone,18 the other three by an independent local doctor6,7⇓). The rest relied on the trial investigators or did not specify.

It became apparent that there were differences in the measure of functional outcome used between trials: some used a “poor functional outcome” and some used a “good outcome.” Five trials sought “dependency” (modified Rankin 3–6 or Barthel 60 or worse) as a measure of poor functional outcome,8,12,17,18,20⇓⇓⇓⇓ and six trials sought good functional outcome6,7,13,14,21⇓⇓⇓⇓ (modified Rankin 0 or 1). The only trials for which the number of patients in individual Rankin groups were not thus far available (and therefore the data shown are for Rankin 2 or worse) are ATLANTIS A7 and PROACT.21

Two trials, both testing streptokinase (SK), stopped early on the advice of their respective data monitoring committees after only about half of the originally intended number of patients had been randomized.12,17⇓ A third trial of SK18 was suspended by its steering committee because of the stopping of other SK trials to examine its interim results after randomizing only one-third of its originally intended number. ATLANTIS A was stopped on publication of the NINDS trial, and continued in modified form as ATLANTIS B,7 which in turn stopped in 1998 after a “futility analysis” prompted by results from the European Cooperative Acute Stroke Study (ECASS) II.14 Therefore, only four of the trials carried out in the 1990s reached their planned targets.6,8,13,14⇓⇓⇓ The rest either stopped early (as above) or did not explicitly state their planned sample size, or were performed only as a prelude to further trials.

Effect of thrombolysis on outcome.

The results are reported for an analysis of all available data from all trials, and for just the trials using rt-PA. Not all data are available from all trials for all outcomes.

Deaths occurring within the first 7 to 10 days (seven trials9,13-15,17-19⇓⇓⇓⇓⇓⇓).

There was a significant excess of early deaths with thrombolysis: 16.6% of those allocated to thrombolytic therapy died compared with 9.8% of those allocated to control (OR 1.85, 95% CI 1.48 to 2.32; 2p < 0.000001). In absolute terms, this represents an increase of 68 (95% CI 44 to 93) early deaths per 1,000 patients treated with thrombolysis. There was borderline significant heterogeneity [χ² 15.05 (df = 7), p < 0.05]. Data on early deaths were available for four trials using i.v. rt-PA13-15,19⇓⇓⇓ and showed a non-significant excess of early deaths (OR 1.24, 95% CI 0.85 to 1.8; 2p = NS) with no significant heterogeneity.

Intracranial hemorrhage (17 trials6-21⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓) (figure 1).

It is difficult to estimate the exact number of symptomatic (or fatal) intracranial hemorrhages because some patients died without a CT scan or postmortem. Therefore, the true number with symptomatic intracranial hemorrhages may be higher than that suggested by these data. The ECASS trial13 did not report symptomatic intracranial hemorrhage but rather the radiologic appearance (hemorrhagic transformation of an infarct or parenchymatous hematoma). Most parenchymatous hematomas were associated with symptoms, so they was used as a surrogate for symptomatic intracranial hemorrhage.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Symptomatic (including fatal) intracranial hemorrhage.

There was a highly significant fourfold increase in symptomatic intracranial hemorrhage with thrombolysis (9.4% of those allocated to thrombolysis versus 2.5% of those allocated to control; OR 3.5, 95% CI 2.8 to 4.5; 2p < 0.000001) with an extra 70 (95% CI 58 to 83) symptomatic intracranial hemorrhages per 1,000 patients treated.

Data on fatal intracranial haemorrhage are available from 11 trials.6,7,10,12-15,17,18,20⇓⇓⇓⇓⇓⇓⇓⇓⇓ There was a significant fivefold increase in the rate of fatal intracranial hemorrhage with thrombolysis (5.4% of patients allocated to thrombolysis versus 1.0% of those allocated to control; OR 4.15, 95% CI 2.96 to 5.84; 2p < 0.000001). There was no heterogeneity (χ² for heterogeneity 8.99; p = NS).

In trials using rt-PA, there were 73 (95% CI 55 to 90) extra symptomatic intracranial hemorrhages per 1,000 patients treated (OR 3.2, 95% CI 2.4 to 4.3; 2p < 0.000001), and there were 29 (95% CI 17 to 41) extra fatal intracranial hemorrhages per 1,000 patients treated (OR 3.2, 95% CI 2.0 to 5.2; 2p < 0.000001) with no heterogeneity between trials.

Total deaths (including early deaths) from all causes by the end of follow-up (17 trials6-21⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓) (figure 2).

There was a modest but significant increase in deaths during follow-up, from 15.9% in controls to 19% in the patients allocated to thrombolysis (OR 1.31, 95% CI 1.13 to 1.52; 2p = 0.0008). In absolute terms, this represented an extra 36 (95% CI 17 to 56) deaths at the end of follow-up per 1,000 patients treated with thrombolysis. There was considerable heterogeneity between the trials [χ² 38.7 (df = 17); p < 0.01].

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Total deaths (including early deaths) from all causes by the end of follow-up.

In the trials using i.v. rt-PA, there was a non-significant increase in deaths (OR 1.16, 95% CI 0.94 to 1.44) equivalent to 18 more deaths per 1,000 patients treated. There was significant heterogeneity of treatment effect among the trials of rt-PA (χ² 14.16 (df = 7), p < 0.05).

Functional outcome (data on 4,476 patients from 12 trials6-8,12-14,17-21⇓⇓⇓⇓⇓⇓⇓⇓⇓⇓) (figure 3).

The latest time of treatment administration in these 12 trials was 6 hours. There was a significant reduction in death or dependency (Rankin 3–6) with thrombolysis: 55.2% of those allocated to thrombolytic therapy compared with 59.7% of those allocated to control (OR 0.83, 95% CI 0.73 to 0.94; 2p = 0.003). This would be clinically important if confirmed, as it is equivalent to 44 (95% CI 15 to 73) fewer dead or dependent patients per 1,000 treated. There was no significant heterogeneity of treatment effect [χ² 19.96 (df = 12); p = NS], i.e., broadly speaking, the treatment effect in all trials was in the same direction. For the six trials using i.v. rt-PA (2,764 patients) from which the data were available,6,7,13,14,19⇓⇓⇓⇓ the odds ratio was 0.79 (95% CI 0.68 to 0.92; 2p = 0.002), equivalent to 57 (95% CI 20 to 93) fewer patients being dead or dependent per 1000 treated. There was significant heterogeneity of treatment effect among the trials using rt-PA [χ² 12.82 (df = 5); p < 0.05].

• Download figure
• Open in new tab
• Download powerpoint

Figure 3. Death or dependency (modified Rankin 3–6) by the end of follow-up. Patients treated up to 6 hours after the stroke.

If an alternative definition of “poor outcome” (Rankin 2–6) is used in this analysis, the overall effect of thrombolysis is unchanged, with a significant reduction in the number of patients with a “poor outcome” (OR 0.79, 95% CI 0.69 to 0.90), with no significant heterogeneity. For the six trials using rt-PA, the odds ratio was 0.76 (95% CI 0.65 to 0.89), i.e., no qualitative change from the analysis by Rankin 3–6, but still with significant heterogeneity [χ² 14.84 (df = 5); p < 0.05].

Identifying possible causes of heterogeneity.

To attempt to identify possible causes for the heterogeneity, further pre-specified subgroup analyses were undertaken to ascertain the effect of (a) thrombolytic drug used, (b) concomitant antithrombotic drug usage on total deaths (because that was the outcome with the most complete data); and (c) time to treatment (on functional outcome and total deaths). There were insufficient data to undertake an analysis of the effect of stroke severity (an individual patient data meta-analysis is required for this). However, it is worth noting that the trial that randomized the greatest proportion of severe strokes (patients unconscious)17 also had the highest mortality rate.

(a) There were no statistically significant differences in treatment effect on total deaths between trials using UK (OR 0.71), SK (OR 1.43), and rt-PA (OR 1.16). The confidence intervals were wide, reflecting the relatively small sample sizes. Therefore it is possible that there are insufficient data as yet to demonstrate a clear difference or that the heterogeneity is due to factors other than the thrombolytic drug used.

(b) It is not possible to comment on the effect of ASA use before the stroke, because the data could not be extracted from most publications. The interaction between thrombolytic drugs and antithrombotic drugs given simultaneously was tested only by random allocation in MAST-I,18 which therefore provides the only reliable evidence. In MAST-I,18 there was a clinically important adverse interaction between ASA and SK when these were given simultaneously. This resulted in a substantial increase in case fatality (early and late), which was not offset by a reduction in the number of dead or dependent patients by the end of follow-up (28% of those allocated to SK alone versus 43% of those allocated to SK plus ASA were dead by the end of follow-up [p < 0.001], and 62% and 63% were dead or dependent respectively [versus 68% in the group who received no ASA or SK]). The increase in death in the SK plus ASA group was largely attributable to intracranial hemorrhage.

Reasonably reliable data are available on antithrombotic drug use in 14 other trials, allowing the trials to be ranked according to antithrombotic drug use. There was a trend toward increased deaths, the more frequent and nearer to the administration of thrombolysis the concomitant antithrombotic drug use (OR 1.93 when all patients received antithrombotic drugs within 24 hours of thrombolysis; 1.27 when some patients received antithrombotic drugs within 24 hours; 1.14 when no patients received antithrombotic drugs within 24 hours but some thereafter; and 0.89 for no antithrombotic drugs within the first 10 to 14 days). Although these data are based mainly on non-random comparisons, they do support the evidence for a clinically significant adverse interaction between thrombolytic and antithrombotic drugs given concurrently found in MAST-I,18 and they may partly explain the heterogeneity between the trials for case fatality.

(c) When we consider the effect of the time to treatment (seven trials6,12-15,17,18⇓⇓⇓⇓⇓⇓) (figure 4), the data are limited and are strongly influenced by the NINDS trial, which contributed more than 50% of the data. There are also likely to be imbalances in baseline variables between the thrombolysis and control patients, as evidenced by the small and uneven numbers of patients. These data should therefore be regarded with extreme caution and require confirmation in future trials. There was a significant reduction in the number of patients dead or dependent with thrombolysis given within 3 hours of the stroke (55.2% of those allocated to thrombolysis were dead or dependent compared with 68.3% of those allocated to control; OR 0.58, 95% CI 0.5 to 0.7; 2p = 0.00002).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4. Death or dependency (modified Rankin 3–6) by the end of follow-up. Patients treated within 3 hours of the stroke.

In absolute terms, if confirmed, this would be equivalent to 126 (95% CI 71 to 181) fewer dead or dependent patients per 1,000 treated with thrombolysis and would be highly important clinically. In trials using rt-PA, the equivalent figure was 140 (95% CI 77 to 203) fewer dead or dependent per 1,000. In patients treated within 3 hours of stroke, there was a very modest, non-significant excess of deaths during follow-up with thrombolysis, of 22.3% of patients allocated to thrombolysis versus 20.7% of those allocated to control (OR 1.11, 95% CI 0.84 to 1.47). If confirmed in future trials, this would be equivalent to 17 extra deaths per 1,000 patients treated with thrombolysis within 3 hours of the stroke (95% CI 28 less to 62 more). In trials using rt-PA, the equivalent figure was 12 fewer deaths per 1,000 (95% CI 61 fewer to 38 more).

Discussion.

Overall, thrombolytic therapy was associated with a significant excess of deaths within the first 7 to 10 days and of symptomatic and fatal intracranial hemorrhages and deaths by the end of follow-up. However, disability was reduced in survivors, so overall there was a significant net benefit. For every 1,000 patients treated with thrombolysis up to 6 hours after stroke, 44 avoided death or dependency. Trials using IV rt-PA contributed the most data to this review, and rt-PA appeared more favorable than other drugs. rt-PA was associated with a non-significant excess of early deaths and deaths by the end of follow-up and with a significant excess of symptomatic intracranial hemorrhages. However, significantly more patients avoided dependent survival. For every 1,000 patients treated with i.v. rt-PA, 57 avoided death or dependence when treated up to 6 hours after stroke, and 140 when treated within 3 hours of the stroke. This result is statistically and clinically highly significant. However, there was significant heterogeneity of treatment effect for total deaths (all trials) and in trials testing rt-PA, on the outcomes “total deaths” and “death or dependency.” This indicates the relative instability of the data and highlights the importance of searching for (a) explanations and (b) more data to better define the precise treatment effect.

The excess of early deaths with thrombolytic therapy appears to be due mainly to intracranial hemorrhage. The combination of antithrombotic drugs with thrombolysis within the first 24 hours of treatment appeared particularly hazardous. However, there is no information on the effect of thrombolysis if patients are taking ASA at the time of their stroke, or on when it might be safe to start ASA after the stroke. Randomization of mainly “severe” strokes may increase case fatality (the hazard) with thrombolytic treatment, although there were insufficient data to examine this. An individual patient data meta-analysis would help, but further trials should closely examine the effect of stroke severity and ASA use.

Although the data in patients treated within 3 hours in other trials support the NINDS conclusion,6 the sample size is tiny and there are imbalances such that the data must at present be regarded with caution. The time window beyond which there is unlikely to be any benefit (or too much hazard) with thrombolytic therapy is unclear. The 3-hour time window is one factor to explain the NINDS trial result, but other possibilities to consider are (a) minor imbalances in baseline stroke severity among the treatment groups, (b) strict avoidance of antithrombotic drugs within 24 hours of rt-PA, (c) rigorous control of the patient’s blood pressure during the treatment infusion, (d) the particular type of hospital setting in which the trial was conducted, or (e) the “play of chance.”22

The subgroup of patients randomized within 3 hours of the stroke in the MAST-I trial18 (patients allocated to SK alone), Australian Streptokinase Trial (ASK),12 ECASS,13 and ECASS II14 showed a reduction in the proportion of dead or dependent patients similar to the NINDS trial. However, this meta-analysis shows that there is significant benefit up to 6 hours after stroke. Therefore, the time window for benefit might extend to, or even beyond, 6 hours in selected patients.

It should be noted that most of the trials (all of the recent rt-PA trials) performed the follow-up at 3 months. The NINDS trial recently published data on functional outcome at 6 months and 1 year, which indicate that the effect of rt-PA was sustained beyond 3 months,23 but there are no other data on whether the benefit of thrombolysis is sustained (or even increases) at 1 year. This information would be important for understanding the impact of thrombolytic treatment on health economics.

There are three obvious methodologic problems that may have influenced these trial results. Only five trials used central telephone randomization, all other methods allowing the possibility of anticipating the next treatment allocation. Only four trials used blinded independent follow-up. Thrombolysis, because of its effects on the coagulation system at high doses, is difficult to blind completely because of the obvious signs of bleeding (e.g., at venipuncture sites, easy bruising, gingival or conjunctival hemorrhages). Therefore, provision of an identical-appearing placebo (in the syringe) may not fully blind investigators to treatment allocation. Follow-up by an independent person who had not been involved at all in the administration of the trial treatment or in the care of the patient during at least the first few days, and was unaware of treatment allocation, is important to ensure objective outcome assessment. Only 4/16 (25%) achieved their planned sample size. Five others stopped early, three were only intended to be small, and four did not state their planned target. Trials that stop early are usually excessively positive or negative because of chance effects,22 and therefore may seriously bias the estimate of treatment effect.

In addition to the heterogeneity among the trials for two of the main outcomes, raising questions about the precise treatment effect, there are many areas in which there are insufficient data on how best to select and treat patients. (a) There is little information on the effect of thrombolytic therapy in the elderly, in whom stroke is most common. Only three recent trials did not have an upper age limit. (b) The presence of a visible recent infarct on the pre-randomization CT scan may be related to increased hazard, but this was based on a post hoc analysis of the CT scans in ECASS,13 in which the baseline CT scans were not read blind to follow-up CT scans. Some trials had CT-visible infarction exclusion criteria and some, including NINDS, did not. The reported rate of CT-visible infarction varied among trials, reflecting either differences in patient selection, observer sensitivity, or definition of the signs of visible infarction. There is no information on the effect of other possible risk factors on the CT scan (such as evidence of “small vessel disease” or previous infarcts). Furthermore, there is no information whatsoever on how other forms of imaging might help guide patient selection (e.g., transcranial Doppler, magnetic resonance, spectroscopy). (c) This review is based on data from just over 5,000 patients, a very small number in relation to the global burden of the disorder (perhaps 6 million ischemic strokes per year worldwide). The centers that took part in the trials were especially interested in, and familiar with, the investigation and management of acute stroke. To extrapolate these results to thrombolysis when used more widely in “routine” clinical practice in less specialized centers could result in much greater hazards and thereby reduce or completely negate any potential benefit.

Therefore, before thrombolytic drugs should be used outside of very strict licensing regulations, much more information is needed about the following: (a) how to select patients (to maximize benefit and minimize hazard); (b) the influence of stroke severity; (c) stroke subtype; (d) age; (e) time from onset; (f) concomitant use of antithrombotic drugs; (g) choice of thrombolytic drug; (h) dose and route of administration; and (i) CT scan findings.

More information could be extracted from the existing data in an individual patient data meta-analysis. However, that will not address the problem of what sort of care environment is required, of patient age, brain imaging, latest time window, or prior ASA use, because there are insufficient data. The existing data raise many questions for which further new data are required to provide concrete answers and for which randomized trials are the best means of obtaining such data.2 For all these reasons, although it may be reasonable to use rt-PA in highly selected patients in experienced centers at present, much more data from further trials are needed to precisely define the absolute treatment effect, which patients to avoid, and which patients will benefit.