Stroke: a clock, a clot, and the brain you can still save

Start here: what an ischemic stroke is

The brain is 2% of body weight and takes about 20% of the body's oxygen, and it stores almost none of it. Interrupt the blood supply to a region of brain and the neurons there begin to die in minutes.

An ischemic stroke is that interruption: an artery supplying the brain is blocked, usually by a clot that either formed on a diseased vessel wall or travelled there from the heart or the neck. Roughly four out of five strokes are ischemic; the rest are haemorrhagic, a burst vessel rather than a blocked one — a distinction that matters enormously, because the treatment for one would kill someone with the other ^[1].

What the blockage does, in two zones. At the centre of the territory, where blood flow falls to almost nothing, tissue dies quickly: this is the infarct core, and it is unsalvageable. Around it sits a rim where flow is reduced but not absent — kept marginally alive by collateral vessels rerouting blood around the blockage. This is the ischemic penumbra: tissue that is not working but is not yet dead, and that will die if flow is not restored ^{[2] [3] [4]}. Everything acute stroke medicine does is an attempt to save the penumbra before it becomes core.

How fast that happens varies enormously between people, and collateral supply is the main reason. Good collaterals buy hours; poor collaterals cost them ^{[3] [5] [6]}. Downstream, the ischemic cascade involves excitotoxicity, blood–brain-barrier breakdown and — paradoxically — further injury when blood returns ^{[7] [8] [9] [10]}.

Scale. Stroke is one of the largest causes of death and disability on earth. In 2019 there were 12.2 million incident strokes, 101 million people living with stroke, 143 million DALYs, and 6.55 million deaths ^{[11] [12]}.

And it is largely preventable. The INTERSTROKE case-control study across 22 countries established the risk-factor profile for ischaemic and haemorrhagic stroke ^[13], and its expanded analysis across 32 countries quantified the global and regional effect of potentially modifiable risk factors ^[14]. Hypertension leads that list, which is why this review's closest sibling is hypertension.

Why urgency is the whole story. Unlike most conditions in this collection, stroke is an emergency measured in minutes, and the reason is quantifiable — it is the subject of the centrepiece below. Three pillars follow: measurements, treatment, and progress.

Pillar 1: measurements and diagnosis

Recognising it: the part that happens outside hospital

Most of the delay in stroke care happens before anyone reaches a doctor, so recognition is a public skill, not just a clinical one. FAST — Face drooping, Arm weakness, Speech difficulty, Time to call emergency services — is the tool built for that. Its diagnostic accuracy has been assessed across primary care, emergency physicians and ambulance staff ^[15], paramedic-recorded FAST signs agree well with physician assessment ^[16], and national awareness campaigns built on it have been formally evaluated ^{[17] [18] [19]}. Prehospital stroke scales are now also designed to do something harder: predict large-vessel occlusion specifically, because those are the patients who need to go to a thrombectomy-capable centre rather than the nearest hospital ^{[20] [21] [22]}.

Grading it: the NIHSS

Once in hospital, severity is scored on the National Institutes of Health Stroke Scale (NIHSS) — a structured neurological examination covering consciousness, gaze, visual fields, facial and limb power, ataxia, sensation, language, speech and neglect, summed to a score where higher means worse ^{[23] [24]}. It does three jobs at once: it quantifies the deficit reproducibly enough for trials, it predicts outcome (combined with age, in scores such as SPAN) ^[25], and a high NIHSS raises the probability of a large-vessel occlusion, which changes where and how the patient is treated.

Outcome is scored differently: the modified Rankin Scale (mRS), 0 (no symptoms) to 6 (death), is the standard endpoint, and "functional independence" is conventionally mRS 0–2 while "excellent outcome" is 0–1 ^{[26] [27] [28]}.

Imaging: three questions in sequence

First: is it a bleed? A non-contrast CT is done immediately, and its primary job is not to show the stroke — early ischemic change is subtle — but to exclude haemorrhage, because thrombolysis into a bleeding brain is catastrophic ^[29].

Second: is there a large-vessel occlusion? CT angiography images the arteries and finds the blockage, and also shows collateral filling, which predicts how fast the penumbra is being lost ^{[5] [4]}.

Third: how much is core and how much is penumbra? CT perfusion measures cerebral blood flow, blood volume and transit times, and the difference between the severely and mildly hypoperfused regions estimates the salvageable mismatch. A prospective multicentre study of 130 patients systematically evaluated which perfusion parameters best predict infarct and penumbra against diffusion-weighted MRI as the gold standard ^{[2] [30] [31]}.

That third question is the one that changed practice. It converts "how long since onset?" into "how much brain is still savable?" — and as the Progress section shows, that substitution is what opened the late treatment windows.

Centerpiece: time is brain, quantified

Everything above serves one purpose: getting blood back into the penumbra before it dies. The cost of delay has been calculated explicitly, and unusually for the models in this collection, every parameter here is grounded — all of them come from a single source.

By combining consensus estimates of the number of neurons in the human forebrain, the volume of a typical large-vessel supratentorial stroke, and the time such a stroke takes to complete, Saver quantified the loss rate. The typical final infarct volume is 54 mL (sensitivity range 19–100 mL); the average duration of nonlacunar stroke evolution is 10 hours (range 6–18); the human forebrain holds 22 billion neurons. From those:

> "In patients experiencing a typical large vessel acute ischemic stroke, 120 million neurons, 830 billion synapses, and 714 km (447 miles) of myelinated fibers are lost each hour. In each minute, 1.9 million neurons, 14 billion synapses, and 12 km (7.5 miles) of myelinated [fibers are lost]" ^[32].

As a model, that is simply a linear loss until the stroke completes:

N_lost(t) = r · min(t, T)

with r = 1.9 million neurons per minute and T ≈ 10 hours. Two consequences follow immediately. A completed stroke destroys about 1.14 billion neurons — roughly 5% of the forebrain. And the neurons spared by faster treatment depend only on the minutes saved: reperfusing at 90 minutes rather than 4.5 hours spares 180 × 1.9 million ≈ 342 million neurons.

One honest note the script enforces rather than hides: the source states both a per-minute and a per-hour figure, and they are rounded inconsistently — 1.9 million/min implies 114 million/hour against the stated 120 million/hour. The computation uses the per-minute figure and asserts the two agree within 10% (they differ by 5%), rather than silently choosing one.

What the model explains. Three things.

First, why stroke systems are built around minutes rather than hours. Every 15 minutes shaved off door-to-needle time spares about 28 million neurons. That is the arithmetic behind pre-notification, bypassing triage, CT-scanner-first protocols and mobile stroke units.

Second, why treatment benefit decays rather than switching off at a deadline. The pooled analysis of individual patient data from 6,756 patients across nine randomised alteplase trials found that treatment delay, age and stroke severity all modify the effect, with good outcome defined as mRS 0–1 ^[33]; the equivalent pooled analysis of five stent-retriever trials characterised the period over which thrombectomy is associated with benefit and how strongly delay erodes it ^{[34] [35] [36]}. The regulatory windows are cutoffs drawn across a continuous decline.

Third — and this is the subtlety that late-window trials exploit — the 10-hour figure is an average, not a law. Its stated range is 6 to 18 hours, and the variation is largely collateral supply ^{[32] [3] [6]}. Patients whose clock runs slowly still have penumbra at 12 or 20 hours. You cannot identify them from the clock; you can identify them from imaging. That single observation is what the Progress section is about.

Limits, honestly. The model describes an average large-vessel supratentorial stroke and says nothing about a small lacunar one. It treats loss as linear when the real trajectory depends on residual flow. Neuron count is a poor proxy for what a person loses — an infarct in eloquent cortex and one of equal volume elsewhere are not equivalent injuries, which is exactly why outcome is scored on the mRS rather than in millilitres ^[26]. And it says nothing about reperfusion injury, which is real ^{[10] [9]}.

Pillar 2: treatment

Two reperfusion strategies, then prevention of the next one.

Intravenous thrombolysis: dissolve the clot

Alteplase is recombinant tissue plasminogen activator — the body's own fibrinolytic trigger, given as a drug. It converts plasminogen to plasmin, which digests the fibrin holding the clot together. Long-term follow-up of the pivotal trial showed benefit persisting at one year ^[37], and ECASS III then extended the window: 821 patients randomised to alteplase or placebo between 3 and 4.5 hours after onset, with disability at 90 days as the primary endpoint, after CT exclusion of haemorrhage or major infarction ^{[29] [38] [39]}. European guidelines now codify the indications ^[40].

The cost is bleeding. Symptomatic intracranial haemorrhage is the principal harm, its risk factors are characterised ^[41], and its management has its own scientific statement ^[42]. Systematic reviews weigh the benefit and harm together ^{[43] [44]}.

Alteplase also has a mechanical limitation the trials exposed: it is poor at opening large arteries. Recanalisation rates with IV rt-PA in large-vessel occlusion are low ^[45], and the site of occlusion predicts whether it will work ^[46]. Hence the second strategy.

Endovascular thrombectomy: pull the clot out

For a large-vessel occlusion — internal carotid or proximal middle cerebral artery — a catheter is advanced from the groin or wrist into the blocked artery and the clot is physically retrieved with a stent retriever or aspirated out ^[47].

2015 was the year this became standard. A cluster of randomised trials reported in quick succession — rapid endovascular treatment ^[48], perfusion-imaging-selected therapy ^[49], stent retriever after IV t-PA versus t-PA alone ^[50] — and the HERMES individual-patient-data meta-analysis of five trials settled it ^[51], with the companion analysis quantifying how benefit falls with time to treatment ^{[34] [52]}.

Open questions have narrowed since. Whether IV thrombolysis should still be given before thrombectomy when both are available has been tested head-to-head ^{[53] [54]}. And thrombectomy has been extended to patients with large established cores who were previously excluded ^{[55] [56]}.

Secondary prevention: stopping the next one

• Antiplatelets. Aspirin irreversibly inhibits cyclo-oxygenase, blocking thromboxane-mediated platelet aggregation; clopidogrel blocks the platelet P2Y12 ADP receptor. The collaborative meta-analyses established antiplatelet therapy for prevention of serious vascular events ^{[57] [58]}. For minor stroke or high-risk TIA, short-course dual therapy is better than aspirin alone: POINT randomised 4,881 patients to clopidogrel plus aspirin or aspirin alone with a 90-day composite of major ischemic events ^[59], and ticagrelor plus aspirin has been tested in the same population ^[60].
• Statins. SPARCL randomised 4,731 patients with recent stroke or TIA and no known coronary disease to atorvastatin 80 mg or placebo; over a median 4.9 years, fatal or non-fatal stroke occurred in 11.2% versus 13.1% (HR 0.84, 95% CI 0.71–0.99) ^{[61] [62] [63]}.
• Blood pressure. The leading modifiable risk factor ^{[13] [14]}, covered in its own right in the hypertension review.
• Anticoagulation for atrial fibrillation. A fibrillating atrium forms clots that embolise to the brain, and these strokes are large. Warfarin worked but was difficult; the direct oral anticoagulants replaced it on the strength of large randomised comparisons — dabigatran, which inhibits thrombin, and the factor Xa inhibitors rivaroxaban and edoxaban, all tested against warfarin ^{[64] [65] [66] [67]}. Antiplatelet therapy is not a substitute for anticoagulation in AF.

Pillar 3: progress

Extended windows, selected by imaging rather than by clock. This is the largest change of the last decade, and it follows directly from the centrepiece's caveat that 10 hours is an average. DAWN enrolled 206 patients with internal carotid or proximal MCA occlusion last known well 6 to 24 hours earlier who had a mismatch between clinical deficit severity and infarct volume ^[68]. DEFUSE 3 enrolled patients 6 to 16 hours out with an initial infarct under 70 mL and a perfusion-to-infarct mismatch ratio of at least 1.8, and was terminated early for efficacy after 182 patients ^{[69] [70]}. For patients who wake with symptoms and have no known onset time, MRI-guided selection using the DWI–FLAIR mismatch has been tested for thrombolysis ^[71].

Tenecteplase is displacing alteplase. It is a bioengineered variant of tPA with greater fibrin specificity and a longer half-life, given as a single bolus rather than a bolus plus one-hour infusion — which matters when the patient must be moved to an angiography suite. It produced better reperfusion before thrombectomy than alteplase ^{[72] [73]}, and non-inferiority has been tested in successively larger pragmatic trials: NOR-TEST ^[74], ATTEST ^[75], the Canadian AcT trial ^[76], and TRACE-2 ^[77], leading to an expedited European recommendation ^{[78] [79]}.

Mobile stroke units take the CT scanner to the patient. A randomised comparison of diagnosis and treatment in a mobile stroke unit versus in hospital ^[80], ambulance-based thrombolysis and its effect on time to treatment ^{[81] [82]}, benefits versus standard management ^[83], and an analysis associating MSU dispatch with functional outcomes ^[84] — plus telemedicine and telestroke in the ambulance ^{[85] [86]} — all attack the same term in the model: minutes before treatment.

Neuroprotection remains the unsolved half. Reperfusion saves the penumbra by restoring flow; nothing yet reliably makes neurons tolerate ischaemia while waiting. NMDA-receptor targeting is the longest-running attempt ^[87], and reperfusion injury is now understood well enough to be a target in its own right ^{[10] [9] [8]}.

And systems keep improving faster than drugs. Endovascular results improve measurably with institutional experience ^[88], workflow analysis shows where the minutes actually go ^[36], and standardised outcome measurement makes the comparisons possible ^[89].