The Extended Window Becomes the Default: Reperfusion 4.5-24 Hours After Stroke

Bottom Line: After three decades in which a clock governed thrombolysis, imaging-based tissue selection has become the organizing principle of late reperfusion. Two 2026 meta-analyses now show that intravenous thrombolysis given beyond 4.5 hours to imaging-selected patients improves excellent functional outcome (pooled relative risks of 1.22 and 1.23; number needed to treat roughly 12 to 13), at the cost of more symptomatic intracranial hemorrhage and with no significant mortality increase. The 2026 AHA/ASA guideline already endorses imaging-selected thrombolysis at 4.5 to 9 hours (Class 2a) and at 4.5 to 24 hours for large-vessel occlusion without thrombectomy access (Class 2b) — and the newest trials extend the case further. Thrombectomy remains first-line for large-vessel occlusion; selection remains everything.

How Time Lost Its Monopoly

The treatment of acute ischemic stroke was built around a clock. From the original 3-hour alteplase window through its extension to 4.5 hours, time from onset served as a surrogate for the only question that matters biologically: is there still salvageable brain? The penumbral concept — a core of irreversibly injured tissue surrounded by hypoperfused but viable brain — implied that some patients keep a favorable tissue profile for many hours, while others lose it within the first one. Advanced imaging made that profile visible, and a decade of randomized trials has now tested whether the tissue clock can replace the wall clock. The answer, first for thrombectomy and now for thrombolysis, is yes — for the right patients.

Thrombectomy Built the Road

Late-window reperfusion was proven first with devices. DAWN randomized patients 6 to 24 hours from last known well with a clinical–core mismatch (small automated core volumes despite substantial deficits) and found thrombectomy superior on utility-weighted mRS at 90 days (5.5 vs 3.4; posterior probability of superiority >0.999) (DAWN, NEJM 2018). DEFUSE 3, using perfusion mismatch at 6 to 16 hours, was stopped early for overwhelming efficacy: functional independence in 45% versus 17% with medical therapy, a number needed to treat of 4 (DEFUSE 3, NEJM 2018). MR CLEAN-LATE then broadened selection to the mere presence of collateral flow on CT angiography at 6 to 24 hours, improving the mRS distribution (adjusted common OR 1.67, 95% CI 1.20–2.32), though with more symptomatic hemorrhage (7% vs 2%) (MR CLEAN-LATE, Lancet 2023). The principle was established: with the right imaging, "late" patients behave like early ones.

Thrombolysis Catches Up — a Winding Path

Proof of concept

WAKE-UP applied an MRI tissue clock — a visible diffusion lesion without corresponding FLAIR change, suggesting onset within roughly 4.5 hours — to patients with unknown onset time. Alteplase increased favorable outcome (mRS 0–1) at 90 days from 41.8% to 53.3% (adjusted OR 1.61, 95% CI 1.09–2.36; number needed to treat 9) (WAKE-UP, NEJM 2018). EXTEND used automated perfusion mismatch (CT perfusion or MR perfusion–diffusion) at 4.5 to 9 hours or on awakening, and alteplase improved excellent outcome (35.4% vs 29.5%; adjusted RR 1.44, 95% CI 1.01–2.06; number needed to treat 17), with more symptomatic hemorrhage and an early stop at 225 of 310 planned patients (EXTEND, NEJM 2019).

Where it failed — and what the failures teach

The negative trials are not noise; they mark the boundaries of the strategy. TWIST selected wake-up strokes with non-contrast CT alone and found no benefit from tenecteplase (adjusted OR 1.18, 95% CI 0.88–1.58) — plain CT does not identify the tissue profile that benefits (TWIST, Lancet Neurology 2023). TIMELESS gave tenecteplase 4.5 to 24 hours after last known well to perfusion-selected large-vessel occlusion patients of whom 77% underwent thrombectomy, and the mRS distribution did not differ (adjusted common OR 1.13, 95% CI 0.82–1.57) — on top of modern endovascular care, additional late thrombolysis has no proven role (TIMELESS, NEJM 2024). ETERNAL-LVO, stopped early for drug supply and a shifting evidence landscape, was numerically unfavorable (primary outcome 37% vs 43%; adjusted RR 0.90, 95% CI 0.66–1.21) with more symptomatic hemorrhage (ETERNAL-LVO, Stroke 2025). ROSE-TNK (n=80) showed early neurological improvement but no 90-day difference (ROSE-TNK, J Stroke 2023), and CHABLIS-T II tripled major reperfusion without symptomatic hemorrhage (33.3% vs 10.8%; adjusted RR 3.0) yet did not change 90-day function (CHABLIS-T II, Stroke 2025). Late thrombolysis is not a universal good; it is a targeted therapy for imaging-selected patients who will not otherwise be reperfused.

The confirmatory wave

Four positive trials in three years settled the question for specific populations. TRACE III randomized large-vessel occlusion patients with salvageable tissue and no access to thrombectomy to tenecteplase at 4.5 to 24 hours: freedom from disability rose from 24.2% to 33.0% (RR 1.37, 95% CI 1.04–1.81; number needed to treat 12) (TRACE III, NEJM 2024). HOPE, in patients 4.5 to 24 hours out with CT-perfusion mismatch and no planned thrombectomy, found alteplase improved mRS 0–1 from 26% to 40% (adjusted RR 1.52; number needed to treat 7), with symptomatic hemorrhage rising from 0.5% to 3.8% (HOPE, JAMA 2025). OPTION extended the paradigm to non-large-vessel stroke: tenecteplase at 4.5 to 24 hours improved excellent outcome from 34.2% to 43.6% (RR 1.28, P=.02; adjusted RR 1.32, P=.007; number needed to treat 11), with a significant increase in symptomatic hemorrhage (2.8% vs 0%, P=.004) (OPTION, JAMA 2026). And TRACE-5 became the first phase 3 trial in confirmed basilar-artery occlusion to show benefit up to 24 hours: mRS 0–1 or return to baseline in 38% versus 29% (adjusted relative rate 1.50, 95% CI 1.09–2.08) without increased symptomatic hemorrhage or mortality (TRACE-5, Lancet 2026).

The Trials at a Glance

EVT = endovascular thrombectomy; LVO = large-vessel occlusion; BAO = basilar-artery occlusion; NNT/NNH = number needed to treat/harm. Symptomatic intracranial hemorrhage (sICH) definitions vary across trials; rates are not directly comparable between rows.

What the Pooled Data Say

The two 2026 syntheses agree to a striking degree despite different trial sets. Patel and colleagues pooled 14 randomized trials (4,174 patients) of thrombolysis beyond the conventional window: excellent functional outcome rose (RR 1.22, 95% CI 1.14–1.31; number needed to treat about 12), good outcome rose (RR 1.12), symptomatic hemorrhage more than doubled (RR 2.44, 95% CI 1.45–4.09; number needed to harm 62), and 90-day mortality was not significantly different (RR 1.13, 95% CI 0.93–1.38). Notably, outcomes did not differ significantly between tenecteplase and alteplase trials, nor between DWI–FLAIR and perfusion-based selection (Patel et al., JAMA Network Open 2026). A separate pooled analysis of 12 trials plus the individual-patient-data pooling of EXTEND, ECASS4-EXTEND, and EPITHET (4,867 patients) found nearly identical efficacy — excellent outcome 40.2% versus 32.5% (RR 1.23, 95% CI 1.15–1.33; number needed to treat 13, with no heterogeneity) — with symptomatic hemorrhage 3.1% versus 1.2% (RR 2.11; number needed to harm 75) and mortality unchanged (RR 1.03, 95% CI 0.89–1.18) (Extended-Window IVT Meta-Analysis, Neurology 2026).

In the accompanying editorial, Rabinstein calls imaging-guided thrombolysis at 4.5 to 24 hours "a new standard" for appropriately selected patients — while stressing two cautions that the trial data themselves impose: the benefit was demonstrated in patients with small cores, large penumbras, and mostly moderate deficits; and intravenous thrombolysis must not substitute for thrombectomy where thrombectomy is available for large-vessel occlusion (Rabinstein, JAMA Network Open 2026).

What Is Current Practice

The 2026 AHA/ASA guideline — issued in January, before OPTION, TRACE-5, and the meta-analyses published mid-year — already moved substantially: thrombolysis at 4.5 to 9 hours from onset — or wake-up stroke within 9 hours of the sleep midpoint — with salvageable tissue on advanced imaging carries a Class 2a recommendation, and thrombolysis at 4.5 to 24 hours for large-vessel occlusion with salvageable tissue and no thrombectomy access carries Class 2b (AHA/ASA 2026 Guideline, Stroke 2026). Against that backdrop, a practical reading of the trial evidence today:

• LVO, thrombectomy available: thrombectomy remains the treatment; late add-on thrombolysis has no proven benefit (TIMELESS; ETERNAL-LVO — largely an early-window trial — was also unsupportive).
• LVO (ICA/M1/M2), no timely thrombectomy access: tenecteplase 4.5–24 h in TRACE III-like patients — salvageable tissue on perfusion imaging, NIHSS 6–25, previously independent (TRACE III; largely Chinese trial populations, a generalizability caveat).
• Non-LVO stroke with CT-perfusion mismatch, 4.5–24 h: tenecteplase per OPTION, or alteplase per HOPE (HOPE enrolled a mixed no-planned-thrombectomy population).
• Wake-up stroke without perfusion imaging: the MRI DWI–FLAIR mismatch route (WAKE-UP); plain CT selection is insufficient (TWIST).
• Basilar occlusion ≤24 h: tenecteplase improved outcomes in TRACE-5 (confirmed BAO, favorable pc-ASPECTS; about half also underwent thrombectomy).
• Counseling the trade-off: roughly one additional excellent recovery per 12 patients treated against one additional symptomatic hemorrhage per 62–75, with no significant mortality increase.

What to Expect Next

Three developments seem probable. First, guideline revisions: OPTION, TRACE-5, and both meta-analyses postdate the January 2026 AHA/ASA document, and the non-LVO and basilar populations they cover sit outside its current extended-window recommendations — the next update has clear upgrades to consider. Second, agent and selection questions: pooled subgroups show no significant difference between tenecteplase and alteplase or between imaging strategies, but no adequately powered head-to-head late-window comparison exists. Third, the access problem: the positive anterior-circulation late-window thrombolysis trials leaned on perfusion imaging, which much of the world lacks (WAKE-UP used MRI signatures; TRACE-5 in basilar occlusion needed only vessel confirmation); whether simpler selection — collateral-based, as MR CLEAN-LATE used for thrombectomy, or clinical-radiological scores — can safely extend the paradigm is the field's most consequential open question, and TWIST is the cautionary tale for shortcuts. Meanwhile the systems-of-care implications are immediate: 24/7 perfusion imaging, revised transfer protocols, and door-to-needle thinking that no longer stops at 4.5 hours.

What the General Neurologist Needs to Know

• The reperfusion window is now defined by imaging, not the clock — up to 24 hours for selected patients.
• For anterior-circulation late thrombolysis, tissue imaging (CT perfusion or MRI) is required — plain CT selection failed (TWIST). Basilar occlusion is the exception: TRACE-5 selected by confirmed occlusion and pc-ASPECTS, not perfusion.
• LVO with thrombectomy available → thrombectomy; do not delay it for thrombolysis decisions.
• Positive phase 3 late-window thrombolysis trials now cover: LVO without thrombectomy access (TRACE III), non-LVO with mismatch (OPTION), mismatch with no planned thrombectomy (HOPE), and basilar occlusion up to 24 h (TRACE-5).
• Expect roughly NNT 12 for excellent recovery versus NNH ~62–75 for symptomatic hemorrhage; no significant mortality increase in either pooled analysis.
• AHA/ASA 2026: Class 2a for imaging-selected thrombolysis at 4.5–9 h or wake-up stroke within 9 h of sleep midpoint; Class 2b for 4.5–24 h LVO without EVT access. Newer trials are not yet incorporated.

Conclusion

The 4.5-hour boundary was never biology; it was the resolution limit of the tools available in 1995. With tissue imaging as the arbiter, thrombolysis joins thrombectomy in the extended window — not for everyone, but for the imaging-selected patient who would otherwise go untreated, the evidence is now consistent, replicated, and pooled. The clock has become a screening question. The scan gives the verdict.