Transient ischaemic attack (TIA): the warning that plays out in hours

The patient is sitting in the waiting room and is perfectly well. He mentions, almost apologising for bringing it up, that last Thursday his right arm went heavy for twenty minutes, and that his speech was slurred. It passed on its own. He called nobody. This article deals with what is decided in those twenty minutes recounted after the event: what a transient ischaemic attack has become since its definition changed, what risk it really carries and on what timescale, what the ABCD2 score is worth and above all what it is not worth, and what the physiotherapist must do who, very often, is the first clinician to hear the story.

• Updated August 2026
• Level clinical summary
• Sources 23 verified references (PMID)
• Reading time about 40 minutes

What is a TIA, and why has its definition changed?

For forty years, TIA was defined by a clock. Today it is defined by an image, and that shift is not a quarrel about nomenclature: it moved part of what used to be called TIA over to the side of established infarction, and it made imaging an urgent act rather than a confirmatory test.

From duration to tissue

The historical definition was convenient and wrong. It said: a focal neurological deficit of vascular origin that resolves completely within 24 hours. The 24-hour threshold had no biological basis; it was set by consensus in the mid-1960s, at a time when the brain could not be looked at.

Diffusion imaging showed what the threshold was hiding. In patients whose signs had all disappeared, and well before the twenty-fourth hour, an infarction was found. In 2009 the American Heart Association and the American Stroke Association therefore redefined TIA as a transient episode of neurological dysfunction caused by focal brain, spinal cord or retinal ischaemia, without acute infarction1. Duration leaves the definition; the absence of a lesion enters it.

The practical consequence is direct and often poorly understood: you cannot state that it was a TIA without imaging. Until the brain has been looked at, a resolved deficit is an ischaemic event of undetermined nature. That is exactly the reasoning that justifies the urgency, and not the other way round.

How many TIAs are actually infarctions

The figures are consistent from one cohort to another. In the SOS-TIA emergency clinic at Bichat hospital, 108 of the 643 patients with confirmed TIA, that is 17 %, had a tissue lesion5. In the international TIAregistry.org, which included 4,789 patients in 21 countries and in which 78.4 % were assessed by a specialist within 24 hours, 33.4 % had an acute cerebral infarction on imaging6.

In other words: between one patient in six and one in three referred for a deficit that has fully recovered already have a scar. The tissue has been lost; only the clinical picture has repaired itself, because the area involved was small or functionally redundant.

The mechanism, in one page

A TIA is a brief interruption of blood flow in an arterial territory. Three main mechanisms account for most cases, and they have neither the same prognosis nor the same treatment.

• Large-artery atherosclerosis, first and foremost at the carotid bifurcation. A plaque ulcerates and throws platelet emboli towards the middle cerebral artery or the ophthalmic artery. It is the mechanism with the highest risk of early recurrence, and it is also the only one that calls for urgent surgery.
• Cardiac embolism, dominated by atrial fibrillation. The thrombus forms in the left atrial appendage and migrates. The treatment is anticoagulation, not antiplatelet therapy, which makes identifying the mechanism decisive.
• Small-vessel disease, or lipohyalinosis, which produces the lacunar syndromes. The risk of early recurrence is lower there, but the long-term burden of disability is not negligible.

In the TIAregistry, 23.2 % of patients had at least one extracranial or intracranial stenosis of 50 % or more, and 10.4 % atrial fibrillation6. Both of those causes can be treated, and treated well: which is precisely what makes delay culpable.

“The typical patient loses 1.9 million neurons for each minute their stroke goes untreated.” (Saver, Stroke, 2006)

That sentence concerns established stroke and not TIA20. But it describes exactly what is at stake in TIA: what is in play is not the episode that has passed, it is the infarction that threatens, and whose occurrence is counted in hours and days.

What is the real risk of stroke after a TIA, and over what period?

This is the question that has to be answered accurately, because it governs everything else. And the right answer is not a single figure: the risk after a TIA depends less on the patient than on what is done with them in the hours that follow.

An average figure that hides considerable heterogeneity

Giles and Rothwell's meta-analysis pooled 18 independent cohorts and 10,126 patients. The pooled risk of stroke at 7 days is 5,2 % (95 % CI 3.9 to 6.5). But the heterogeneity between studies is massive (p less than 0.0001), with risks ranging from 0 % to 12.8 %3.

The substance of the article lies in explaining that spread. The authors show that it is almost entirely explained by method, setting and treatment, and not by the patients:

A factor of twelve separates the two extremes. That is not a difference in severity between patients, it is a difference between care systems. The authors' conclusion is sober and weighty: the lowest risk is seen where treatment is given urgently in a specialist service.

The two demonstrations that changed practice

Two studies published in the same year, 2007, turned a correlation into a demonstration.

EXPRESS compared two consecutive periods in the same Oxford population. In phase 1 the patient was seen in clinic and preventive treatment was then started by the family doctor; in phase 2 the assessment and the first prescription took place the same day in clinic. The median time to assessment fell from 3 days to under 1 day, and the median time to the first prescription from 20 days to 1 day. The risk of recurrent stroke at 90 days fell from 10.3 % to 2.1 %, an adjusted hazard ratio of 0.20 (95 % CI 0.08 to 0.49; p equal to 0.0001)4. Two points deserve emphasis: the patients' baseline characteristics and the time they took to seek care were comparable between the two phases, and early treatment did not increase the risk of intracerebral haemorrhage or other bleeding.

SOS-TIA, in Paris, opened a clinic accessible round the clock and circulated its number to 15,000 doctors across the Île-de-France region. Of 1,085 patients admitted, 74 % went home the same day, and the stroke rate at 90 days was 1,24 % (95 % CI 0.72 to 2.12) when the rate predicted by the ABCD2 scores was 5.96 %5. The median symptom duration there was 15 minutes.

The modern risk, and the risk that does not go away

Contemporary cohorts, managed in organised pathways, find figures markedly lower than those of the 2000s. TIAregistry.org gives the following Kaplan-Meier estimates after a TIA or minor stroke: 1.5 % at 2 days, 2.1 % at 7 days, 2.8 % at 30 days, 3.7 % at 90 days and 5.1 % at 1 year6.

The five-year follow-up of the same registry, covering 3,847 patients, brings the most useful information for the physiotherapist, who follows these patients for a long time. The cumulative stroke rate is 9,5 % at 5 years (95 % CI 8.5 to 10.5), and above all 43.2 % of those strokes occurred between the second and the fifth year7. The rate of cardiovascular events is 6.4 % in the first year, and another 6.4 % over the following four.

This result bears directly on everyday practice. The patient who had a TIA two years ago is not a cured patient: they remain, for a long time, a patient at high vascular risk. Their TIA belongs in their physiotherapy history, whatever they are consulting for today.

How do you recognise a TIA, and what is not one?

This is the difficult part, and it should be said plainly: there is no diagnostic test for TIA. The diagnosis rests entirely on a story, often reported after the event, by a patient who has nothing left to show. The rate of wrong diagnoses is consequently high, in both directions.

The clinical signature of an ischaemic event

Three features, taken together, point strongly to a vascular origin.

• The deficit is focal. It corresponds to an arterial territory, and therefore to an anatomy. A general faint turn, diffuse fatigue, tingling in all four limbs are not focal.
• Onset is sudden and the deficit is maximal from the start. Ischaemia does not spread step by step: it strikes a territory all at once. A march of symptoms over several minutes argues rather for another cause.
• The symptoms are negative. They are losses of function: loss of strength, loss of sensation, loss of speech, loss of vision. Positive phenomena (shimmering, tingling that spreads, involuntary movements) point rather to a migraine or an epileptic seizure.

The two main territories

What the FAST mnemonic catches, and what it misses

The Face Arm Speech Test was built for prehospital alerting and it does that job well. In Harbison's study, ambulance staff trained in FAST made a correct diagnosis in 144 of the 183 stroke patients who presented to them, that is 79 %, and they brought markedly more patients in within 3 hours than general practitioners did (46 % against 12 %)12.

But FAST has a documented blind spot. Aroor and colleagues reviewed 736 consecutive records of ischaemic stroke: 14.1 % had no FAST symptom on admission. Among them, 42 % had a balance disturbance or lower limb weakness, 40 % visual symptoms, and 70 % one or the other. Adding those two items, so that the mnemonic becomes BE-FAST (Balance, Eyes, Face, Arm, Speech, Time), the proportion of unidentified strokes falls from 14.1 % to 4.4 % (p less than 0.0001)13.

For the physiotherapist, the implication is immediate: a sudden balance disturbance and a sudden visual disturbance are stroke symptoms, just as much as a drooping mouth. They are also, precisely, the complaints for which a patient may end up referred for vestibular rehabilitation or balance work rather than to the emergency department.

The mimics: what looks like a TIA without being one

The review by Nadarajan and colleagues, in Practical Neurology, sets the problem out well: the transience of the symptoms makes the differential diagnosis harder than that of established stroke, and the rate of mimics there is higher and more varied10. In Wardlaw's meta-analysis, 45 % of patients referred to a stroke prevention service were mimics9.

What is the ABCD2 score worth, and above all what is it not worth?

This is probably the point on which the gap between what is taught and what has been demonstrated is widest. The ABCD2 score still features in many teaching materials as the triage tool for TIA. The prospective data, however, say that it does not sort reliably, and above all that it does not see what matters.

What the score says

The score was derived and validated by Johnston, Rothwell and colleagues in 2007, by unifying two earlier scores (the California score and the ABCD score), across four independent cohorts totalling 2,893 patients2.

In the original publication, the stratification of stroke risk at 2 days was clear: 8.1 % for scores 6 and 7 (21 % of patients), 4.1 % for scores 4 and 5 (45 %), and 1.0 % for scores 0 to 3 (34 %)2.

What the score does not do

Two later pieces of work, one prospective and one meta-analytic, seriously dented the confidence that could be placed in it.

Perry's prospective validation enrolled 2,056 patients in eight Canadian emergency departments, with the score calculated by the physician before any decision about disposal. The observed stroke rate was 1.8 % at 7 days and 3.2 % at 90 days. A score above 5 had a sensitivity of 31.6 % (95 % CI 19.1 to 47.5) for stroke at 7 days: two strokes in three occurred in patients that threshold classified as low risk. A score above 2, the threshold recommended by the American Heart Association, did reach a sensitivity of 94.7 %, but with a specificity of 12.5 %, that is, at the price of classifying almost everyone as at risk. The area under the curve was 0.56 when the score was calculated by the front-line physician. The authors conclude that the score is inaccurate, whatever threshold is used8.

Wardlaw's meta-analysis, across 29 studies and 13,766 patients, adds the heaviest criticism. A score of 4 or more is sensitive (86.7 %) but not very specific (35.4 %) for recurrence at 7 days. And above all:

• 20 % of patients with a score below 4 had carotid stenosis of more than 50 % or atrial fibrillation, that is, a cause requiring urgent intervention;
• 35 to 41 % of mimics had a score of 4 or more, against 66 % of true TIAs: the score therefore does not separate TIAs from what is not one either;
• out of 1,000 patients referred, 52 % would score 4 or more, which does not reduce the clinic's workload9.

An analysis of the same TIAregistry had already shown that patients with an ABCD2 below 4 could have a 90-day stroke risk comparable to that of patients scoring 4 or more, as soon as they had large-artery atherosclerosis21.

What predicts better than the clinical score

Imaging does what the clinical picture does not. Coutts and colleagues followed 120 patients with TIA or minor stroke, examined within 12 hours and imaged by MRI within 24 hours. The risk of stroke at 90 days, adjusted for baseline characteristics, was 4.3 % with no diffusion lesion, 10.8 % with a lesion but no arterial occlusion, and 32.6 % with both a lesion and an occlusion (p equal to 0.02)11. A factor of seven separates the extremes, in clinically similar patients.

It is that observation which led to scores incorporating imaging, such as ABCD2-I, where adding the brain and the carotids improves the identification of patients at early risk22. The general lesson does not change: what stratifies is investigation, and investigation is done urgently.

What happens in the pathway, and why do the hours count?

Knowing what awaits the patient you refer is not idle curiosity: it is what lets you explain to them why they cannot wait until Monday, and convince them, which is often the hardest part.

The work-up, and what it looks for

Assessing a TIA aims to answer a single question: which mechanism, and therefore which treatment. It classically comprises brain imaging, investigation of the cervical and intracranial arteries, a search for a rhythm disturbance, and metabolic blood tests. The 2021 American guideline for stroke prevention sets out how it is organised16.

The search for atrial fibrillation deserves a separate mention, because it is the cause whose diagnostic yield depends most on the length of recording, and because its treatment differs radically. A patient labelled with a “cryptogenic TIA” on antiplatelet therapy may be a patient in paroxysmal atrial fibrillation who should have been anticoagulated.

Where that care takes place is not a matter of indifference. The Cochrane review of stroke units, which pooled 29 trials and 5,902 participants, shows that organised care in a dedicated unit reduces, compared with a conventional ward, the risk of a poor outcome (odds ratio 0.77; 95 % CI 0.69 to 0.87), of death (0.76; 0.66 to 0.88) and of death or dependency (0.75; 0.66 to 0.85), with moderate certainty evidence23. That is one more argument for referral to the stroke pathway where one exists, and not to a general service.

Two treatments whose benefit runs out with time

Short-course dual antiplatelet therapy. The POINT trial randomised 4,881 patients with minor ischaemic stroke or high-risk TIA between clopidogrel plus aspirin and aspirin alone. Major ischaemic events occurred in 5.0 % of patients on dual therapy against 6.5 % on aspirin alone (hazard ratio 0.75; 95 % CI 0.59 to 0.95; p equal to 0.02), most events occurring during the first week. The price is an increase in major haemorrhage, 0.9 % against 0.4 % (hazard ratio 2.32; 95 % CI 1.10 to 4.87)15. The CHANCE trial, conducted in China on the same principle, had opened this path14. The window of efficacy is narrow, which is one more argument against the deferred appointment.

Carotid endarterectomy. This is probably the most striking illustration of the cost of delay. Rothwell's pooled analysis, across 5,893 patients and 33,000 patient-years of follow-up, shows that the benefit of surgery depends not only on the degree of stenosis but on the time since the last event. For a stenosis of 50 % or more, the number of patients needing surgery to prevent one ipsilateral stroke at 5 years is 5 if the operation takes place within 2 weeks, and 125 beyond 12 weeks17.

The benefit was also greater in men, and in patients aged 75 and over, which contradicts the instinct to rule older patients out.

The physiotherapist sees the patient afterwards: what should they do?

Here is the heart of the subject for our profession. The physiotherapist is not the one who makes the diagnosis of TIA nor the one who investigates it. They are very often the one who hears the story first, because they see the patient over time, regularly, and in a setting where people talk. That position gives them no additional diagnostic competence; it gives them a responsibility to refer, and that alone.

Why the physiotherapist is on the front line in spite of themselves

Three structural reasons, worth naming.

The resolution of symptoms defuses the alarm. A patient whose arm stayed heavy for twenty minutes and who has been well since has no natural reason to think emergency. So they mention it to whoever they see, which is often their physiotherapist, in passing during a session booked for something else.

Posterior circulation symptoms look like reasons for rehabilitation. A sudden balance disturbance, vertigo, unsteadiness when walking: these are common reasons for referral. The patient may arrive at the clinic with a prescription for vestibular rehabilitation or balance work for what was an ischaemic event. That is exactly the blind spot BE-FAST put figures on13.

The frequency of sessions makes the physiotherapist a sensor. Seeing a patient three times a week for six weeks is thirty chances to hear “by the way, the other day...”. No other clinician has that density of contact.

The three situations, and what to do in each

What can physiotherapy do in secondary prevention?

Two things must be clearly distinguished here: what physical activity contributes as a modifiable risk factor, which is firmly established at population level, and what a supervised rehabilitation programme contributes after a TIA, which is far less so. Confusing the two would be selling a certainty that does not exist.

The frame: ten risk factors explain most of it

The INTERSTROKE study, across 32 countries, establishes that ten potentially modifiable risk factors are collectively associated with about 90 % of the attributable risk of stroke, in every major region of the world, in men as in women and at every age. Hypertension is more strongly associated there with intracerebral haemorrhage, whereas current smoking, diabetes, apolipoproteins and cardiac causes are more strongly associated with ischaemic stroke19.

Physical activity is one of those factors, and it is the only one the physiotherapist acts on directly in their practice. That is the rational basis for their involvement, and it is solid.

What a recent trial measured, and what it did not find

Honesty requires the following to be reported. Walton and colleagues published in 2026 in Clinical Rehabilitation a randomised controlled trial of cardiovascular rehabilitation after TIA or minor stroke: 140 participants, median age 73, randomised between a weekly six-week programme and usual care, with assessment at 6 weeks and 6 months.

The primary result is negative. No difference between groups on six-minute walk test distance in the adjusted analysis. The only significant differences concerned one dimension of quality of life: the AQoL-6D coping score at 6 weeks, in favour of rehabilitation (beta 3.6; 95 % CI 0.07 to 7.2). No other outcome differed. In subgroup analysis, the participants with a TIA proper showed a non-significant gain in walking (Cohen's d 0.40). The authors conclude that cardiovascular rehabilitation was no more effective than usual care in improving exercise capacity18.

What remains defensible in practice

What do concrete clinical situations teach us?

The three situations that follow are fictional cases, built to illustrate a line of reasoning. Every figure they draw on refers back to its source; the patients themselves do not exist.

How do you apply this in the clinic?

What to tell the patient, and how

Convincing them is the real difficulty. The patient feels well, they have things to do, and the idea of going to the emergency department for a symptom that has gone strikes them as disproportionate. Three ways of putting it work better than an instruction.

• “What happened to you is not the disease, it is the warning.” The patient grasps the idea of a fire alarm better than that of a probabilistic risk.
• “The fact that it passed on its own does not mean it has been repaired. In some patients, the scan finds a lesion.” That is true and sourced: between 17 % and 33.4 % depending on the cohort56.
• “If we find the cause today, we divide the risk by five. In a fortnight, that will no longer be true.” That is a faithful translation of EXPRESS and of Rothwell's analysis of surgical timing417.

Frequent mistakes

Frequently asked questions

Further reading on the site

This article deals with the transient event and its urgency. For established stroke, the site offers two complementary entries: stroke as a disease, which sets out recognition, thrombolysis, thrombectomy and secondary prevention, and rehabilitation after a stroke, which covers the recovery trajectory over months. On the differential diagnoses met in the clinic, see cerebellar ataxia for the rapid onset of a cerebellar syndrome, vestibular neuritis and benign paroxysmal positional vertigo for acute vertigo, and normal pressure hydrocephalus for gait disturbance in older people.