Best Clinical Practice

Controversies in Transient Ischemic Attack Evaluation and Disposition in the Emergency Department

Brit Long, MD; Alex Koyfman, MD


J Emerg Med. 2017;52(3):299-310. 

In This Article


Assessment of patients with suspected TIA should be conducted in a rapid manner, as stroke and TIA must be differentiated in consideration of thrombolytic therapy. This evaluation of TIA can also determine which patients require admission or discharge. Several factors associated with higher stroke risk include age over 60 years, infarct discovered on imaging, cardiogenic emboli, and modified Rankin score > 2.[1,2,17] Hypertension, diabetes, and coronary artery disease do not demonstrate statistically significant associations, though trends are present.[16–18] The evaluation of suspected TIA centers on neuroimaging and the use of clinical risk scores, with the goal of risk stratification. However, the specific imaging required in the ED and patient disposition based on risk scores are controversial topics.


The AHA/American Stroke Association (ASA) recommends neuroimaging within 24 h of suspected TIA. Magnetic resonance imaging (MRI) with diffusion-weighted imaging (DWI) is the preferred imaging modality, but computed tomography (CT) is most commonly available in the ED, as 56% to 92% of patients receive imaging with this modality in the ED.[3–6,8,19]

Head CT. Rapid neuroimaging is utilized in patients with focal deficits and concern for ischemic stroke, as thrombolytics require the absence of hemorrhage. Head CT noncontrast is the primary imaging modality, which can rapidly identify other conditions such as intracranial hemorrhage or mass. This imaging modality displays sensitivities ranging from 12% for small emboli in peripheral vascular distributions to 52% for large proximal occlusions.[3–6,8,17,19–22] Förster et al. in 2012 found that 95.7% of initial head CT examinations were negative for acute infarction.[19] A study from Germany evaluating head CT noncontrast in 1533 patients with suspected TIA found a 3.1% rate of acute cerebrovascular accident, despite complete resolution of symptoms.[20] Of the 1533 patients, 17 suffered ischemic stroke. However, no patients with new infarct on imaging experienced stroke while in the hospital.[20] Another study in 2003 found the frequency of stroke did not differ at 90 days in those receiving head CT vs. those who do not.[21] This same study did endorse the use of CT to evaluate for other etiologies of symptoms, as 1.2% of patients possessed an alternative condition found on head CT.[21] These lesions included chronic subdural hematoma and mass. Head CT noncontrast is not reliable for acute ischemia, but it can find alternative conditions necessitating management.

MRI. Literature supports the use of MRI in acute ischemic stroke and TIA evaluation, specifically the use of MRI with DWI. This modality receives a Class I, Level B recommendation for patients with suspected TIA.[6,8] DWI will demonstrate hyperintense signals due to restricted diffusion with cytotoxic edema during the acute phase.[23–25] Close to one-third of patients with normal CT and MRI noncontrast demonstrate acute lesions on DWI.[23] Approximately 39% of patients possess ischemic lesions on imaging, and follow-up scanning past 24 h reveals involvement in up to 100% of patients.[24,25] These ischemic lesions found on MRI, even in the setting of complete symptom resolution, predict future stroke, with 15-fold increase in stroke.[26–28]

This test may not be available in the ED, as this is the first-line modality in 5% to 15% of centers, with MRI available for 40% of cases.[29,30] MRI displays greater diagnostic capabilities for ischemic lesions than CT, as 35.2% of patients with negative CT may display ischemic lesions on MRI.[19] Within 12 h of acute stroke symptom onset, MRI with DWI demonstrated an odds ratio of 25 (95% confidence interval [CI] 8–79) if ischemia is discovered with this modality, whereas another study found an odds ratio of 10.1 for acute stroke within 7 days, with acute ischemia diagnosed on MRI with DWI[23,24,27,28] Sensitivity ranges from 83% to 97% for early ischemic lesions.[31–33]

Stroke risk in negative DWI ranges from 0 to 2.9% at 2 days and 7 days, whereas patients with scans positive for ischemia possess a stroke rate of 14.3% at 2 days and 23.8% at 7 days.[23,27,31–33] Intermediate- to high-risk scores from clinical rules are not associated with abnormalities found on DWI.[34–36] Clinical symptoms such as unilateral weakness or facial palsy are correlated.[37]

Patients with positive DWI remain at high risk for stroke, no matter the predicted risk on clinical scoring. A separate study by Calvet et al. included 339 patients undergoing MRI with DWI for suspected TIA.[25] DWI is positive in 40% of these patients, and factors associated with positive imaging included weakness, duration of symptoms > 60 min, atrial fibrillation, and large artery atherosclerosis.[28] These findings have been replicated in several other studies, where time of symptoms and focal weakness predict true ischemic lesion. Negative MRI with DWI is associated with low risk of stroke, especially when used in conjunction with risk stratification[25,27,28,38] A study by Asimos et al. reported MRI performed in 1168 patients for TIA, in which 331 patients had positive MRI findings.[38] Based on this study's results, patients with negative MRI and low ABCD2 are at extremely low risk for stroke.[38] This score will be discussed.

Vascular Imaging. A major risk for stroke and recurrent TIA includes significant carotid stenosis, defined by occlusion > 70% and > 50% with symptoms in males.[5] The AHA/ASA provides a Class 1, Level A recommendation for intracranial and extracranial vascular imaging in the routine evaluation of suspected TIA.[3–5,8] Up to 31% of patients with TIA have carotid disease, and in the setting of significant disease, 90-day stroke risk can reach 20.1%.[39–41] The presence of carotid disease in association with ABCD2 score has been evaluated, and stroke rate is highest in patients with large artery atherosclerosis, which includes carotid disease. However, higher scores on ABCD2 rule do not necessarily predict carotid disease.[37,41–43] Carotid disease alone is a significant risk factor for adverse outcome including recurrent stroke, as a study by Calvet et al. found a hazard ratio of 4.9 (95% CI 1.4–16.9, p = 0.006).[25] Importantly, close to half of patients with positive lesions on DWI have significant stenosis of at least one large intra-/extracranial vessel.[25]

Treatment for carotid disease with carotid endarterectomy has been shown effective for high-grade disease, defined by stenosis > 70%.[44,45] Cerebral angiography was the gold standard for diagnosis, but this modality is invasive, requiring arterial cannulation. New, noninvasive modalities including carotid ultrasound (US), computed tomography angiography (CTA), and magnetic resonance angiography (MRA) demonstrate efficacy. D'Onofrio et al. evaluated carotid Doppler US, digital subtraction angiography, and MRA.[46] Based on this study, these tests are able to diagnose surgical lesions requiring treatment with sensitivities > 95%.[46] Negative likelihood ratio for MRA and US is 0.07.[47] Another study evaluating patients with TIA found a sensitivity of 88% and specificity of 76% for carotid stenosis by US.[48] MRA displays a sensitivity of 92.2% and specificity of 75.7%.[46–49] These results have been supported in several other studies, with sensitivities of Doppler US approaching 80% and MRA 95%.[49,50] Meta-analyses demonstrate sensitivities of 82% to 86% and specificities of 90% to 100% for significant disease > 70% occlusion.[51–54] One meta-analysis compared Doppler US, MRA, and CTA.[54] Sensitivities for Doppler US, MRA, and CTA were 89%, 94%, and 77%, respectively, with specificities 84%, 93%, and 95%.[54] Literature suggests that stenosis < 50% on Doppler US or MRA is associated with low likelihood of significant disease. However, stenosis > 50% requires further imaging with MRA or CTA.[3–6,8,51–54]

If available, Doppler US and MRA possess adequate sensitivity and specificity for diagnosis of significant carotid disease. CTA is likely easier to obtain in most EDs, but this test alone may miss significant disease.

Atrial Fibrillation

Atrial fibrillation is a major risk factor for stroke, independent of imaging and risk prediction tools.[3–6,8,55] In fact, close to 2% of patients with TIA will be diagnosed with new-onset atrial fibrillation.[55,56] However, risk scores, particularly the ABCD and ABCD2 scores, are not correlated with atrial fibrillation.[57] The diagnosis and acute management of atrial fibrillation, including anticoagulation, may reduce short- and long-term risk of stroke.[3–6,8]

Risk Scores

Immediate evaluation is needed for patients with TIA or stroke to determine if an indication for thrombolytics is present, while also categorizing patients into separate categories for those requiring admission vs. discharge.[1–6,8,10,26,28,35,37,57,58] A significant percentage of patients with TIA are at risk for stroke after the acute event, and investigators have sought to derive and validate risk-stratification tools that predict the risk of stroke after TIA.[1–6,8,10,26,28,35,37,57,58] Goals of using risk-stratification tools include identifying patients at low risk for whom further work-up may be deferred, while identifying patients at short- and long-term risk of stroke who may receive benefit from further evaluation and neurology consultation.[1,2] Several risk-stratification tools are present, shown in Table 2.[1–6,8,10,26,28,35,37,57,58] Unfortunately, all of the scores were evaluated in patient populations with low rate of stroke, poor blinding of investigators and patients, and a significant rate of patients lost to follow-up.

One of the first evaluations for stroke risk, known as the ABCD score, uses age (1 point), blood pressure > 140/90 mm Hg (1 point), unilateral weakness (2 points), speech impairment without weakness (1 point), and symptom duration > 60 min (2 points) or duration 10–59 min (1 point) as predictors of stroke, with increasing risk as the number of points increases.[1,16,22,35–37,58–66] Patients with a score of 0–3 are considered low risk, whereas those > 3 points are considered moderate to high risk. Low-risk scores demonstrate 2-day, 7-day, 30-day, and 90-day risks of 1.2%, 5.9%, 5.4%, and 3.2%, respectively, with the moderate- to high-risk patients demonstrating risks of 4.9–7.9%, 4.2–15.9%, 6.9–17.6%, and 11.3–18.9%, respectively.[1,22,60–66] The California rule is similar to the ABCD score. However, it does not have hypertension but does use diabetes.[1,22,35,58,60,61] Both the ABCD and California scores categorize over 54% to 85% of the studied populations as at least intermediate risk.[1,16,22,35–37,58–66] This is a major weakness when attempting to find patients appropriate for discharge. Area under the curve (AUC) of the receiver operating characteristic measures a prognostic score's discriminatory capability, with scores ranging from 0.5 to 1.0. A score < 0.75 possesses fair accuracy, 0.75–0.92 good accuracy, and above 0.92 very good accuracy.[28,58,60,61,63–70] The ABCD and California scores demonstrate an AUC of 0.62 to 0.81, with the majority of studies demonstrating values of ≤ 0.70.[28,58,69] This value correlates with fair accuracy for predicting stroke in these patients.

The most commonly used and evaluated tool is the ABCD2 score. This score adds diabetes to the original ABCD score. Initial studies validating this score suggest strong predictive attributes for stroke risk at 24 h.[58,60,64,65,69] Investigators first used 1916 patients in the derivation group, followed by 2892 patients in the validation group. Using this score for stratification, 33%, 48%, and 19% are categorized as low, moderate, and high risk, respectively.[27,28,35,58,60,61,63–75] The score demonstrates sensitivities of 86% in moderate- to high-risk patients, though specificity is 35%. Close to 1% of this group experience stroke at 2 days, with 1.2% at 7 days. The AUC ranges from 0.66–0.74 for 2- and 7-day stroke risk. Initial results show stroke occurs in 3.2% of patients at 90 days. However, positive likelihood ratios never reach higher than 1.54.[28,35,58,63–75]

Many providers in the past have relied upon these scores. However, they demonstrate limited predictive ability. Schrock et al. in 2009 suggested that high-risk ABCD2 score is not beneficial for guidance on obtaining other diagnostic testing, including MRI, electrocardiogram (ECG), and head CT.[58,69] Only positive carotid duplex study demonstrated correlation with high-risk scores.[58,69] Perry et al. in 2011 suggested that it is not a reliable tool, as a cutoff of 5 points resulted in misclassification of approximately 8% of patients as low risk.[75] This cutoff displayed a sensitivity of 94.7%, but a specificity of 12%.[75] Stead et al. in 2011 found no difference between different classifications based on the ABCD2 score, as the low-, moderate-, and high-risk groups display stroke rates of 1.1%, 0.3%, and 2.7% at 7 days and 2.1%, 2.1%, and 3.6% at 90 days, respectively.[76] A study by Ghia et al. in 2012 found stroke rates in low-risk ABCD2 patients to be 1.2% at 30 days and 0.8% in moderate- and high-risk groups, questioning ability for risk stratification and stroke prediction.[77]

Other validations have questioned the use of this score. A study in an Australian population suggests patients in all risk categories possess similar stroke rates, while at the same time having poor predictive ability.[28,58,65,77] When used in combination with other imaging modalities evaluating the brain and carotid systems, the score does not provide additional risk-stratification information. Sensitivity in high-risk patients may approximate only 30% to 40%.[65,69,70,71,75,76,77] Schrock et al. suggest the use of this test alone misses patients with high-grade carotid stenosis.[69] A 2012 meta-analysis of 33 studies found a positive likelihood ratio of 1.4 for scores > 3, with sensitivities of 89% at days 2 and 7 and 87% at day 90 post TIA.[65] Authors suggest this score does not have predictive capability likely to change management in the ED.[65]

Investigators have sought to improve the ABCD2 score by adding imaging. The ABCD2-I score added CT or MRI with DWI, which results in an AUC value of 0.78 at 7 days, vs. 0.66 for the original ABCD2 score.[67] A modification to the score created the ABCD3-I score, with the third "D" representing a TIA occurring within 1 week of the first TIA.[74] The "I" component refers to carotid stenosis > 50% discovered on carotid imaging or any abnormality discovered on MRI with DWI. This score does not demonstrate moderate predictive capabilities, though it does demonstrate better ability when compared with the original ABCD2 score.[28,58,74] The C-statistic for the modified score is 0.66, whereas the ABCD2 score demonstrates a C-statistic of 0.54, neither over the threshold of 0.7 for moderate prediction. When imaging involves MRI with DWI, this value reaches 0.81, demonstrating greater prognostic predictive value than ABCD2 score.[28,58,74,75]

Another more recently developed risk score was developed in Canada. Investigators enrolled 3906 patients, with strokes occurring in 2.2%.[78] Over 97% of the cohort underwent initial head CT in the ED. They found several factors associated with stroke, from which they created the Canadian TIA score. The score is shown in Table 3, with scores ranging from −3 to 23 and stroke rate within 7 days ranging from 0.01% to > 27%.[78] Patients with < 6 points demonstrate < 1% chance of stroke, with sensitivity approaching 98%. Scores > 10 demonstrate 5.1% stroke risk, with scores > 12 possessing a 12.6% risk. The discriminatory capability of this test possesses a C-statistic of 0.77 (95% CI 0.73–0.82).[78] However, this score requires multiple variables and has not been validated.

Role of Risk Scores

The use of prediction scores alone for risk stratification is not recommended. The predictive capabilities are not reliable, nor are they reproducible.[58,65,69,75,77,79] Over 40% of patients with > 4 on the ABCD2 score are experiencing mimic, whereas 21% of patients with low scores are experiencing symptoms due to atrial fibrillation or carotid stenosis.[37] Scores do not allow recognition of stroke subtype such as lacunar, cardioembolic, or large vessel; nor do they predict the specific vascular territory affected.[58,65,69,75,77,79] MRI with DWI and clinical features may predict risk, independent of risk scores. Cucchiara et al. found scores 0–3 to still have significant risk of stroke (up to 20%).[37] Importantly, close to one-third of patients in the ED are not categorized appropriately into low, intermediate, or high risk.[76,79,80] Ultimately, risk scores are a tool that may assist in gauging short-term risk of stroke, but this should not take precedence over physician gestalt. Risk scores provide no insight into the etiology of the TIA, and the etiology is often targeted for secondary prevention.[4–8,58]

The combination of MRI with risk stratification demonstrates promise and significantly improves the diagnostic and predictive values. The addition of MRI with DWI to the ABCD2 score possesses a higher 7-day stroke risk prognostic ability after TIA.[26–28,38,58] One study demonstrates the absence of lesion on MRI with DWI and ABCD < 4 reaches 100% sensitivity for excluding stroke at 7 days, whereas those with infarction on imaging show a 20-fold increase in stroke risk.[38]

ED-Directed Protocols and Observation Units

ED diagnostic protocols and observation units display promise in reducing length of stay and total cost while improving patient compliance with AHA- and NSA-recommended treatments including antiplatelet and antihypertensive treatment.[3–8,58,80–85] Rapid evaluation units or clinics have been used and studied in Europe. These clinics function on the premise of faster evaluation of TIA patients and reduction in stroke rate, as opposed to hospital admission. Studies demonstrate faster time to risk stratification and treatment, as well as a significant reduction in stroke from 10% to approximately 1–2% with use of these clinics.[85–87]

Stead et al. evaluated TIA patients in an ED unit, with the use of a standardized protocol including patients with no symptoms and negative head CT noncontrast.[80] This study found approximately 30% of patients can be discharged directly from an observation unit, with no difference in rate of strokes at 2 and 7 days.[80] Ross et al. in 2007 evaluated 149 patients with suspected TIA in the ED with a diagnostic protocol.[85] Patients with abnormal CT, known embolic source, carotid stenosis, prior stroke, and crescendo TIA were excluded from the protocol. Investigators utilized carotid imaging, echocardiography, repeat neurologic examination, and cardiac monitoring for a period of at least 12 h.[85] No increase in adverse outcomes is present in those patients in the protocol vs. patients admitted to the hospital. Length of stay is shorter (25.6 h; 95% CI 21.9–28.7 vs. 61.2 h; 95% CI 41.6–92.2), as is total cost at 90 days ($890, 95% CI $768–$1510 vs. $1547, 95% CI $1091–$2473).[85] Oostema et al. investigated an ED observation unit that combined the use of MRI with DWI and a diagnostic protocol.[84] Inclusion criteria required the absence of symptoms, as well as no suspected cardioembolic source. However, no CT was completed in the ED, which differs from the majority of ED evaluations. Instead of CT, 94% of patients underwent MRI with DWI, and 97% of patients in the accelerated protocol underwent imaging of the cervical vessels. Close to 14% of patients have infarct on DWI, and these patients demonstrate a 6.3% risk of stroke at 30 days, compared with 1.2% in patients with negative DWI.[84] Length of stay in the diagnostic protocol is 19 h, which is similar to the study by Ross et al..[84,85]

A different model using outpatient clinic is an option as well. Mijalski and Silver in the OTTAWA trial obtained ECG and head CT in the ED, followed by carotid Doppler, echocardiogram, 24-h telemetry, and neurology follow-up (at 7, 14, or > 14 days based on ABCD2 score).[87] Based on these results, the 2-day stroke rate with use of this clinic is 1%, with a 3.2% risk at 90 days. Those patients randomized to ABCD2 score alone demonstrate a 90-day stroke rate of 9.2%.[87] Lavallée et al. found a 90-day stroke rate of 1.24% in patients managed in a hospital-based clinic staffed with neurologists, with imaging including MRI or head CT, carotid ultrasound, ECG, and ankle-brachial index.[81,88] Use of this clinic with consultation and imaging found a stroke rate within 90 days of 1.6%. Of note, 74% of patients can be evaluated and discharged upon presentation to the ED with the use of this clinic, where patients are directly referred.[88] Olivot et al. discharged patients with ABCD2 scores of 0–3 to an outpatient TIA clinic, whereas patients with scores of 4 or 5 underwent imaging of the intracranial and carotid vasculature.[89] If patients demonstrated stenosis > 50% or had scores > 5, they were admitted. Approximately 70% of patients can be discharged from the ED in this study to follow up at the TIA clinic, and stroke rate is lower when compared with the ABCD2 score used alone.[89] Wasserman et al. evaluated 982 patients, with 32% categorized as low risk, 49% as medium risk, and 19% as high risk.[90] All patients underwent head CT and ECG in the ED and follow-up care in a stroke clinic where they received carotid Doppler, echocardiogram, and laboratory testing. Overall stroke risk is 3.2% at 90 days, with < 1% risk for those with scores 0–4. Use of the ABCD2 score predicted a risk of 9.2%.[90] These results have been replicated in other studies, with two from Europe where no difference in stroke risk is discovered between patients admitted and those managed as outpatients in a specialized TIA clinic.[91–94]

The ability to reduce stroke risk, hospital length of stay, and health care costs has been repeatedly demonstrated.[90,93,94] Stroke rate at 90 days can be reduced by 80% with the use of these diagnostic protocols or dedicated clinics.[90,93,94] A specialized diagnostic protocol or a TIA clinic may be equivalent to admission in recurrent stroke rate and treatment. This requires an ED system with resources available, including a protocol or TIA clinic.[58]

Nuts and Bolts for Emergency Providers

The American College of Emergency Physicians published a Clinical Policy on TIA in 2016.[58] A summary of the American College of Emergency Physicians recommendations is shown in Table 4. Guidelines have sought to use risk scores to predict the need for admission. The 2009 AHA guidelines state that patients with ABCD2 score > 2, those with inability to follow up, or those with focal ischemia should be admitted.[3–8] As discussed, risk scores do not have adequate ability to predict low risk vs. high risk.

Patients often require management of long-term, modifiable risk factors such as hypertension, atrial fibrillation, hyperlipidemia, diabetes, and coronary artery disease. Acute evaluation of the patient with TIA is targeted at identification of these risk factors, as treatment of these comorbidities can reduce the risk of stroke for these patients.[3–8,58] As discussed, patients are often admitted for further evaluation and management.

Where does this leave the emergency provider? A proposed algorithm is demonstrated in Figure 1. Ultimately, patients should be evaluated within 24 h from the time of event, whether as an inpatient, in an ED observation unit/diagnostic protocol, or in a specialized outpatient TIA clinic. In the ED, a focused history on the distribution and duration of symptoms is recommended. A detailed and accurate history is imperative, as misdiagnosis by emergency providers occurs in close to 60% of cases. Discordance in diagnosis with neurologists ranges from 42% to 86%.[12,13,28,95,96] The provider should assess for focal neurologic symptoms. Negative symptoms, or those that are associated with loss of function, such as motor weakness, altered speech, or vision abnormalities, suggest TIA, whereas positive symptoms including tingling, increased speech, involuntary motions, and flashing lights, suggest alternative diagnoses.[11,12,13,97] Risk factors including diabetes, hypertension, atrial fibrillation, prior stroke, prior neurologic deficits, hyperlipidemia, and coronary artery disease should be ascertained. An ECG should be obtained to evaluate for atrial fibrillation, which is associated with stroke risk. Neuroimaging is recommended, and per the AHA/ASA, MRI with DWI is the first-line modality.[3–8] However, in most EDs, head CT noncontrast is rapidly available at all times. Any focal lesion found on neuroimaging warrants admission.[2–8,28,58]

Figure 1.

Evaluation of suspected TIA in the ED. ED = emergency department; ECG = electrocardiogram; MRI = magnetic resonance imaging; DWI = diffusion-weighted imaging; CT = computed tomography; TIA = transient ischemic attack.

Patient assessment and availability of local resources will determine the disposition: inpatient admission, ED-focused diagnostic protocol/observation unit, or rapid follow-up in dedicated clinic. Criteria for inpatient admission include crescendo neurologic symptoms or continued symptoms, atrial fibrillation on ECG, vascular disease on imaging, ischemic focus on neuroimaging, poor social situation, inability to follow up, and poor compliance.[2–8,28,58]

If these are not present, a rapid diagnostic protocol or rapid follow-up clinic can be beneficial, which often includes serial examination, telemetry monitoring, MRI and vascular imaging, echocardiography, and specialist consultation for risk factor modification. Specifically, MRI with DWI and carotid imaging are cornerstones of evaluation. Evaluation with these studies should occur within 24 h. ED-focused diagnostic protocols and rapid follow-up clinics decrease patient cost while also decreasing risk of stroke. Stratification tools such as the ABCD, California, ABCD2, ABCD3, or ABCD3-I scores may be used in conjunction with neuroimaging such as MRI with DWI, but these scores alone do not sufficiently identify patients at low risk, whether short-term or long-term, for stroke.