Although the licence restricted the use of rt-PA to within 3 h, the results of the trials⁶ suggested that rt-PA, given as late as 6 h after ischaemic stroke, might reduce the number of patients with poor functional outcomes.^4,5 In that context, the Third International Stroke Trial (IST-3) was initiated in 2000, with the aim of providing evidence from a randomised trial for a wide range of patients, including those older than 80 years, treated between 3–6 h, and with comorbid disorders such as previous stroke. Meanwhile, a meta-analysis of individual patient data from completed trials undertaken in 2004 suggested that intravenous rt-PA might be of benefit up to 4·5 h after stroke,⁶ with the findings confirmed in 2008 in the ECASS 3 trial of patients younger than 80 years who were treated between 3·0–4·5 h after stroke.^7,8 Despite this increasing evidence from 11 trials of 3977 patients who were treated up to 6 h after stroke,⁹ substantial uncertainties¹⁰ have been caused by the lack of precision in the overall estimates of treatment effects for the important outcomes of death and functional outcome, longest time window for net benefit, effect of treatment in people older than 80 years (only 94 patients in this age group were included in the completed trials), and the effect of factors such as stroke severity, stroke subtype, use of antiplatelet drugs before stroke, and other comorbidities (eg, previous stroke, diabetes, or hypertension).

We included all randomised trials of rt-PA versus control in patients with definite ischaemic stroke, after having excluded intracerebral haemorrhage and other brain disorders on imaging that might account for the symptoms. Two reviewers independently assessed study quality, applied the inclusion criteria, extracted the data, and resolved discrepancies by discussion or asking the trial investigators. We excluded trials that did not provide relevant clinical outcomes or in which the latest follow-up was less than 1 month after treatment. We aimed to use intention-to-treat data, sought additional unpublished data when necessary, and verified the extracted data with the principal investigators of the trials when possible, avoiding double counting of data when multiple publications arose from the same trial.

We extracted data using a standard form for trial characteristics including inclusion and exclusion criteria; method of randomisation; latest time to treatment; rt-PA dose and administration; other medical treatments; methods and timing of follow-up assessments; and the demographic details of patients. We extracted data for the number of patients allocated rt-PA or control with key early or late outcomes (detailed below) including the numbers within prespecified categories of modified Rankin Scale (mRS) score, Oxford Handicap Score (OHS, a further modification of the mRS), or equivalent by the end of follow-up. We treated mRS and OHS scale points as being equivalent and translated Barthel Index scores to mRS: Barthel scores greater than 60 are equivalent to mRS 0–2 and Barthel scores greater than 90 are equivalent to mRS 0–1.⁸ We sought to extract data for outcomes in key subgroups defined by delays to treatment (0–3 h, 3–6 h); age (up to and including 80 years or >80 years); presence or absence of visible infarction on imaging; baseline stroke severity; and different antithrombotic regimens. Tabulations of IST-3 data were provided by the trial statistician.

We assessed early outcomes (within 7 days) including death from any cause; fatal intracranial haemorrhage (death attributed to intracranial haemorrhage identified on scanning or at post mortem); death not attributable to intracranial haemorrhage; symptomatic intracranial haemorrhage (as defined in individual trials, but usually as worsening of neurological status contemporaneous with the appearance of new haemorrhage on brain imaging that was sufficient to cause neurological deterioration); and symptomatic infarct swelling (worsening of neurological status contemporaneous with major infarct swelling on imaging). We assessed outcomes at the end of follow-up including death between 7 days and the end of follow-up; total number of deaths from all causes; alive and favourable outcome (defined as mRS or OHS 0–1); alive and independent (defined as mRS or OHS 0–2); and dependency (defined as mRS or OHS 3–5).

Statistical analyses were all prespecified in the review protocol.⁹ The primary analysis was for all patients treated up to 6 h after stroke. We also assessed the effect of rt-PA given within 3 h and between 3–6 h of the stroke, and in patients aged up to 80 years versus those older than 80 years; for analyses subdivided by time to treatment we have used time to randomisation in IST-3 because this provides an unbiased estimate of time to treatment, which is applicable to patients allocated rt-PA and open control. We calculated odds ratios (ORs) and 95% CI for each outcome, unadjusted for baseline variables, using standard fixed-effects methods,¹² and calculated statistical heterogeneity using the χ² statistic, mentioned when significant heterogeneity was present. We provided data for the comparison of the overall estimate from the 11 previous trials with IST-3 and with the updated overall estimate of effect. We calculated the absolute effects per 1000 patients treated. The figures were prepared in R (version 2.11.1).

The effects of rt-PA on the early outcomes are shown in figure 1 and the absolute event rates per 1000 patients treated with estimates of heterogeneity are provided in table 2. Data for deaths within 7 days, available from eight trials,^{2,3,7,13,15,19–21} showed that 8·9% of patients allocated rt-PA and 6·4% allocated control died within 7 days (figure 1), an absolute increase of 25 deaths per 1000 individuals treated (table 2).

Data for fatal intracranial haemorrhage within 7 days, available from eight trials,^{1–3,7,15–18,21} showed that 3·6% of patients allocated rt-PA and 0·6% of those allocated the control died from intracranial haemorrhage (figure 1), an absolute increase of 29 deaths per 1000 patients.

There was no evidence that rt-PA increased the number of deaths within 7 days from causes other than intracranial haemorrhage (figure 1), implying that the excess of early deaths with rt-PA is mostly from fatal intracranial haemorrhage, although it should be noted that data were only available from five trials.^2,3,7,15,21 Although no between-trial heterogeneity was noted (table 2), there was marginal heterogeneity between previous trials as a group and IST-3 (p=0·04; figure 1).

Data for symptomatic intracranial haemorrhage within 7 days, available from all 12 trials, showed that 7·7% of patients allocated rt-PA and 1·8% allocated control developed this haemorrhage (figure 1; appendix p 5). The crude estimate of the absolute excess of symptomatic intracranial haemorrhage with rt-PA was 58 per 1000 patients treated (table 2).

Data for symptomatic infarct swelling within 7 days was available from six trials;^{1–3,7,17,18,21} 10·2% allocated rt-PA and 10·4% allocated control developed infarct swelling (figure 1), with substantial between-trial heterogeneity (table 2) mostly due to differences between IST-3 and previous trials (p=0·001 figure 1).

The effects on outcomes at final follow-up are shown in figure 2 and the absolute event rates in table 2. Data for deaths between 7 days and the end of follow-up, available from eight trials,^{2,3,7,13,15,19–21} showed that 11·5% of patients allocated rt-PA and 13·6% allocated control died in this period (figure 2), a reduction of 22 deaths per 1000 individuals treated with rt-PA (table 2).

Data for total number of deaths from all causes by the end of follow-up, available from all 12 trials, showed that 19·1% allocated rt-PA and 18·5% allocated control died (figure 2; appendix p 5), equivalent to a non-significant increase in deaths of seven per 1000 patients treated with rt-PA.

Data for patients who were alive and independent (mRS 0–2) were available from ten trials (appendix p 6).^{1–3,7,13,16–21} 46·3% of patients allocated rt-PA and 42·1% allocated control were alive and independent (mRS 0–2) at the end of follow-up (figure 2), an absolute increase of 42 per 1000 people treated with rt-PA (table 2).

Data for patients who were alive with favourable outcome (mRS 0–1) were available in the same ten trials (appendix p 6). 34·8% of patients allocated rt-PA and 29·3% allocated control had mRS 0–1 at the end of follow-up (figure 2), equivalent to 55 more patients alive and almost symptom free per 1000 treated (table 2). Heterogeneity was noted between all trials (table 2) but not between previous trials combined and IST-3 (figure 2).

Data for patients treated within 3 h and between 3–6 h after stroke were available for three key outcomes (figure 3, table 2). Among patients treated within 3 h of stroke, using a cutoff of mRS 0–1 (six trials^{1–3,16,18,21}), 283 (31·6%) of 896 allocated rt-PA and 202 (22·9%) of 883 allocated control achieved a favourable outcome (OR 1·61, 95% CI 1·30–1·99, p<0·0001), an absolute increase of 87 per 1000 individuals given rt-PA. Among patients treated within 3 h of stroke (six trials^{1–3,16,18,21}), 40·7% allocated rt-PA and 31·7% allocated control achieved mRS 0–2 (figure 3), an absolute increase of 90 per 1000 patients treated (table 2). Of the patients treated between 3–6 h after stroke (seven trials^{2,3,7,16,17,20,21}), 47·5% allocated rt-PA and 45·7% allocated control achieved mRS 0–2 (figure 2), an absolute increase of 18 per 1000 patients treated. The difference in ORs between the subgroups treated within 3 h and between 3–6 h was significant (χ²=9·49, 2 degrees of freedom (df), p=0·002). There were slightly fewer deaths by the end of follow-up in people treated within 3 h (seven trials;^{1–3,15,16,18,21} figure 3), equivalent to 15 fewer deaths per 1000 (table 2), but slightly more deaths in those treated between 3–6 h (seven trials;^{2,3,7,16,17,20,21} figure 3), equivalent to an excess of 18 per 1000 individuals who were treated with rt-PA. However, the difference in ORs between the subgroups treated within 3 h and between 3–6 h was not significant (χ²=3·20, df 2, p=0·07). No difference was noted in the summary ORs for symptomatic intracranial haemorrhage for those treated within 3 h (six trials;^{1–3,16,18,21} figure 3), equivalent to 68 more such haemorrhages per 1000 treated individuals (table 2), and for those treated between 3–6 h (seven trials;^{2,3,7,16,17,20,21} figure 3), equivalent to 58 more per 1000 treated patients (table 2; test for difference between subgroups χ²=0·57, df 2, p=0·45). The results were the same when the analysis was restricted to the trials of treatment in both time windows (data not shown). No heterogeneity was noted between trials, or between previous trials combined and IST-3, for any of the outcomes within 3 h and between 3–6 h.

Three trials included a total of 1711 patients older than 80 years (25 from EPITHET, 69 from NINDS, the rest from IST-3).^1,20,21 Among patients older than 80 years who were treated within 6 h, 27·2% allocated rt-PA and 23·4% allocated control were mRS 0–2 (p=0·07; figure 4), an increase of 38 per 1000 individuals were alive and independent (table 2). Among patients aged 80 years and younger, 52·5% allocated rt-PA and 48·3% allocated control were alive and independent (mRS 0–2) at the end of follow-up (p=0·009), an increase of 43 per 1000 people (table 2). For patients older than 80 years treated within 3 h, 28·9% allocated rt-PA and 19·3% allocated control were alive and independent (figure 4; p=0·003), an increase of 96 per 1000 people (table 2). Among those 80 years and younger treated within 3 h, 49·6% allocated rt-PA and 40·1% allocated control were alive and independent (figure 4; p=0·001), an increase of 95 per 1000 people (table 2). Among people treated between 3–6 h after stroke, of those aged 80 years and younger 52·6% allocated rt-PA and 50·3% allocated control achieved mRS 0–2 at the end of follow-up compared with 25·9% allocated rt-PA and 26·4% allocated control who were older than 80 years (figure 4).

Do the effects of rt-PA in the 3035 patients in IST-3 (95% of whom did not meet contemporaneous European licence criteria) differ materially from the effects in the 3977 patients from 11 previous trials (who largely did meet these criteria)? The characteristics of patients in IST-3 that excluded them from the licence criteria were chiefly age and time: 53% were older than 80 years, 72% were treated more than 3 h after stroke onset. A large proportion had severe strokes and 146 (5%) had National Institutes of Health Stroke Scale greater than 25, hence the fairly high numbers of deaths in both the rt-PA and control groups. There was no evidence to suggest that any of the beneficial or adverse effects of rt-PA were qualitatively different in these outside-licence patients from individuals in earlier trials, except for non-intracranial-haemorrhage early deaths and symptomatic infarct swelling. The proportion of patients with symptomatic infarct swelling in IST-3 (3·6%) was less than that of previous trials (17·3%) but the results of IST-3 showed an increase in symptomatic infarct swelling with rt-PA, whereas those of previous trials had suggested a reduction (p=0·06). These differences could be attributable to several factors: the analysis of symptomatic oedema and its definition were prespecified in IST-3 but not in five of 11 previous trials that provided relevant data; patients in IST-3 were older, had more severe strokes (hence larger infarcts), and were treated later than in previous trials. These and other possibly aggregate characteristics could have changed the effect of rt-PA on symptomatic infarct swelling and in turn the effect on non-intracranial-haemorrhage early deaths; these points merit further analysis.

The updated review provides additional insights. First, the finding that fewer deaths occurred with rt-PA between 7 days and late follow-up in IST-3 confirms a pattern suspected in previous trials, and indicates that the excess hazard with rt-PA is largely attributable to symptomatic intracranial haemorrhage and occurs early, so that patients who survive the acute phase without intracranial haemorrhage benefit. One explanation for this might be that more disabled stroke survivors have an ongoing higher risk of death;²² rt-PA, by shifting the balance towards better functional outcome, reduces the risk of death in the long term. The longer follow-up in IST-3 (6 months vs 3 months in previous trials), might have allowed this net balance of early hazard versus later benefit to emerge more clearly. Second, the benefits of rt-PA occur in severe and milder strokes. Third, as suggested by the results of non-randomised studies,^23,24 the relative and absolute benefits of rt-PA are at least as large in older as in younger people (figure 4, table 2). Fourth, the new data reinforce the evidence that benefit is greatest when rt-PA is given within 3 h. Therefore, efforts to reduce treatment delays should continue, particularly for older people.

The latest time window for benefit remains unclear. Although benefit from thrombolysis clearly declines with increasing delay to treatment,⁸ these data suggest that benefit probably extends beyond 4·5 h, possibly as late as 6 h in some patients, though the time probably varies with key individual or combined patients' characteristics, which were not possible to identify with just summary unadjusted data. Patients presenting later are different (eg, less severe stroke) than those presenting earlier. The lower benefit with later treatment is not due to more symptomatic intracranial haemorrhage because the odds of this haemorrhage with rt-PA are actually slightly lower in patients treated between 3–6 h than within 3 h (figure 3, table 2). Thus, non-significantly more deaths and the reduction in the OR for alive and independent at the end of follow-up for treatment between 3–6 h might indicate less benefit with later treatment—ie, less tissue to salvage, rather than simply more hazard.