MRI Guides Diagnostic Approach for Ischaemic Stroke

M A Kumar; H Vangala; D C Tong; D M Campbell; A Balgude; I Eyngorn; A S Beraud; J M Olivot; A W Hsia; R A Bernstein; C A Wijman; M G Lansberg; M Mlynash; S Hamilton; M E Moseley; G W Albers

Background and aim Identification of ischaemic stroke subtype currently relies on clinical evaluation supported by various diagnostic studies. The authors sought to determine whether specific diffusion-weighted MRI (DWI) patterns could reliably guide the subsequent work-up for patients presenting with acute ischaemic stroke symptoms.
Methods 273 consecutive patients with acute ischaemic stroke symptoms were enrolled in this prospective, observational, single-centre NIH-sponsored study. Electrocardiogram, non-contrast head CT, brain MRI, head and neck magnetic resonance angiography (MRA) and transoesophageal echocardiography were performed in this prespecified order. Stroke neurologists determined TOAST (Trial of Org 10172 in Acute Stroke Treatment) classification on admission and on discharge. Initial TOAST stroke subtypes were compared with the final TOAST subtype. If the final subtype differed from the initial assessment, the diagnostic test deemed the principal determinant of change was recorded. These principal determinants of change were compared between a CT-based and an MRI-based classification schema.
Results Among patients with a thromboembolic DWI pattern, transoesophageal echocardiography was the principal determinant of diagnostic change in 8.8% versus 0% for the small vessel group and 1.7% for the other group (p<0.01). Among patients with the combination of a thromboembolic pattern on MRI and a negative cervical MRA, transoesophageal echocardiography led to a change in diagnosis in 12.1%. There was no significant difference between groups using a CT-based scheme.
Conclusions DWI patterns appear to predict stroke aetiologies better than conventional methods. The study data suggest an MRI-based diagnostic algorithm that can potentially obviate the need for echocardiography in one-third of stroke patients and may limit the number of secondary extracranial vascular imaging studies to approximately 10%.

Introduction

The accurate diagnosis of stroke subtype is important for the acute management of cerebral ischaemia and for the prognosis and secondary prevention of stroke. Currently, a variety of diagnostic tests exist for the evaluation of acute ischaemic stroke patients. A sequential choice of diagnostic tests should be made to optimise diagnostic yield, reduce the risk of harm to the patient and minimise cost. Challenges may arise when the practitioner is limited by either availability of diagnostic methods or cost-containment policies.^[1] The introduction of powerful diagnostic imaging modalities and aggressive therapeutic strategies requires the need to streamline diagnostic protocols while maintaining standards of care that are well founded, cost-effective and readily available. This has improved accuracy but increased the complexity of clinical decision-making.

Conflicting data exist regarding the diagnostic utility of hospital-based testing in the evaluation of acute ischaemic stroke. The comprehensive in-hospital evaluation of ischaemic stroke patients is performed to accurately determine stroke aetiology. Presumably, knowledge of stroke aetiology leads to better treatment of patients and improved prevention of recurrence. For example, there is clear benefit in evaluating patients for risk factors such as carotid stenosis and atrial fibrillation, which have a distinct treatment path. However, some believe that the standard diagnostic studies do not yield results that warrant prolonged hospitalisation.^[2]

We tested the hypothesis that specific diffusion-weighted MRI (DWI) patterns can more efficiently and accurately identify stroke aetiologies than conventional diagnostic methods. Our aim was to establish whether a comprehensive MRI evaluation has the potential to limit the need for other diagnostic studies and provide a more cost-effective approach to stroke diagnosis. We sought to determine whether MRI could be reliably used as the pivotal study to guide the subsequent diagnostic approach for patients presenting with symptoms of acute ischaemic stroke.

Materials and Methods

Two hundred and seventy-three consecutive adult patients admitted within 48 h of onset of ischaemic stroke symptoms and able to comply with MRI and transoesophageal echocardiography were included in this prospective NIH-sponsored study. All patients, or their designated representative, signed an informed consent to participate in this study. Patients who were intoxicated, who had a severe terminal illness, HIV/AIDS, symptoms referable to an alternative diagnosis such as migraine, seizure or psychiatric disorder, who were in coma or from whom consent was not obtainable were excluded. Patients with a known history of atrial fibrillation were excluded, as their stroke aetiology was presumed to be self-evident; patients who subsequently developed atrial fibrillation were enrolled. Patients who received intravenous tissue plasminogen activator were excluded, as we postulated that thrombolytics could influence the appearance of DWI lesions.^[3] Patients enrolled in other investigational drug studies were also excluded. Transient ischaemic attacks (TIAs) were defined as clinical symptoms suggestive of stroke lasting less than 24 h. The attending stroke neurologist determined a modified Trial of Org 10172 in Acute Stroke Treatment (TOAST) classification on both admission and discharge.^[4]

Patients were to undergo CT, MRI, intracranial and cervical magnetic resonance angiography (MRA) and transoesophageal echocardiography, in this prespecified order. Not all patients underwent all diagnostic testing. Two hundred and fifteen (78.7%) patients had a cervical MRA. Two hundred and forty-one (88.3%) patients had an echocardiogram during their hospitalisation and 196 (81.3%) had transoesophageal echocardiography. No statistically significant difference existed between the rates of acquisition of MRI, MRA or transoesophageal echocardiography for all three groups. All patients had an MRI and MRA of the brain. MRI scans were acquired using the 1.5 GE Signa Horizon scanners (GE Medical Systems, Waukesha, Wisconsin, USA). The MRI sequences that were obtained included: spin echo-echoplanar imaging DWI, fast spin echo, fast spin echo fluid attenuated inversion recovery, gradient echo and multiplanar gradient echo. Apparent diffusion coefficient maps were created from the DWI. Contrast-enhanced MRA (ce-MRA) and three-dimensional time-of-flight MRA (tof-MRA) of the intracranial cerebral circulation and cervical internal carotid arteries were obtained.

The attending stroke neurologist classified each patient’s presumed stroke subtype according to TOAST criteria on admission, after initial clinical assessment (history, examination and baseline laboratory studies) and review of the admission head CT (initial TOAST). The initial TOAST classification was determined prior to acquisition of the MRI scan. The attending stroke neurologist performed the same classification on hospital discharge after all diagnostic tests were completed and final diagnostic assessments were made (final TOAST).

A CT-based classification scheme was compared with an MRI-based one. The initial TOAST subtype was converted to a CT-based classification schema as follows: initial TOAST subtypes 1 and 2 (large artery and cardioembolic) were designated CT-based thromboembolic, initial TOAST group 3 (small-vessel disease) was designated CT-based small-vessel disease and the remaining TOAST groups 4 and 5 (other determined or undetermined) were designated CT-based other. The DWI pattern of the MRI was classified into one of three patterns: (1) thromboembolic, (2) small vessel or (3) other. The thromboembolic pattern was defined as either single or multiple acute ischaemic lesions, in a unilateral or bilateral distribution. Single lesions had to measure greater than 1.5 cm in diameter. The small-vessel pattern was defined as a single small (<1.5 cm) lesion in the subcortical territory or brainstem.^[5] The third category comprised the DWI lesions that did not meet any of the criteria above, including MRIs without a positive diffusion-weighted lesion. The yield of diagnostic tests was compared for both the CT-based and the MRI-based categories. The primary aim of the study was to determine whether the CT-based classification scheme or the MRI-based scheme could better guide the choice of diagnostic studies to predict the reference standard in stroke subtype determination, the physician-determined final TOAST classification.

After review of the DWI scans, the stroke neurologist reviewed the remainder of the MRI, intracranial MRA and cervical MRA scans in that order. Cervical MRA was considered positive if there was a >50% stenosis (North American Symptomatic Carotid Endarterectomy Trial criteria) ipsilateral to a symptomatic hemisphere. The interstudy agreement in the per cent stenosis measurement from the tof-MRA, ce-MRA and carotid ultrasound was compared using Cohen’s κ statistic.

Attending stroke physicians chose one diagnostic test: MRI, MRA, transoesophageal echocardiography, other test or ‘none’ as the principal determinant of change if the final TOAST classification differed from the initial TOAST classification. Other tests included: ECG/telemetry, EEG, catheter angiogram, transcranial Doppler ultrasound, ophthalmological examination, clinical examination, and serological tests of hypercoagulability and peripheral blood smear. If two tests were equally likely to result in the diagnosis, then physicians were to choose ‘none’ as the principal determinant of change. When the initial and final TOAST classes were the same, no test was considered the principal determinant, as no change was made. The tests that constituted the ‘principal determinant of change’ were evaluated for both CT-based and MRI-based approaches. Proportions of the groups were compared using χ² or Fisher exact tests. All tests were two-tailed and significant at level α<0.05. For quality control, one author (MAK) reviewed all cases for consistency. In the event of a discrepancy, the attending stroke physician responsible for determinations was queried.

Results

The demographic characteristics of the cohort are detailed in Table 1. One hundred and seventy-three patients had positive lesions on DWI, compatible with acute ischaemia. Correspondence between the initial and final TOAST classification is shown for the entire cohort in Table 2. The sensitivity and positive predictive value of the initial TOAST score were 60.8% and 47.0% for large artery atherosclerosis, respectively, 40.4% and 67.9% for cardioembolism and 94.0% and 61.8% for small-vessel disease.

One hundred and fifty-three (56.0%) patients had the same diagnosis on discharge as they had on admission. Of the 120 patients who had a change in their diagnosis, transoesophageal echocardiography was the principal determinant of change in 11%, MRI in 13% and MRA in 16% (Table 3). Of the 13 patients for whom transoesophageal echocardiography was the principal determinant of change, 3 had high-risk lesions (left atrial appendage thrombus, aortic valve vegetation and left ventricular thrombus), 6 had medium-risk lesions (5 patent foramen ovale (PFO) and atrial septal aneurysm and 1 mobile atheromatous plaque of the aortic arch) and 4 had low-risk lesions (1 left atrial spontaneous echogenic contrast and PFO, 1 PFO and 2 regional wall motion abnormalities). Of the 19 patients for whom MRA was the principal determinant of change, 11 had culprit lesions in the cervical vessels, 6 had intracranial lesions and 2 had evidence of both. For 55 patients, no test was identified as the principal determinant of change. Of those, eight had more than one possible stroke aetiology. Thirty-three of the 55 patients had no lesion on DWI and a negative diagnostic evaluation.

Stroke is a pathologically heterogeneous disease. A comprehensive approach to the determination of stroke aetiology may be prohibitively costly; alternatively, a cursory evaluation may not adequately prevent recurrence. Our data support the use of a new a diagnostic algorithm for the evaluation of patients presenting with an acute stroke (figure 3). Algorithms can reduce practice variation, limit delays in care, curtail hospital length of stay and minimise patient morbidity.

The proposed model includes initial stratification based on pattern of DWI abnormality after acquisition of the brain MRI/MRA. In the model, the DWI pattern is classified into one of three groups: thromboembolic (includes a unilateral or bilateral distribution of acute ischaemic lesion), small-vessel disease or other. Patients in the thromboembolic category are further stratified according to the presence or absence of ipsilateral arterial atherosclerosis. Our results indicate that the utility of transoesophageal echocardiography is particularly high in the select group of stroke patients who are considered to have a thromboembolic MRI pattern and negative cervical MRA and particularly low in patients with the small-vessel pattern. This approach may obviate the need for transoesophageal echocardiography in patients with a presumed diagnosis of a small-vessel (lacunar) stroke following MRI/MRA.

This study demonstrates a poor correlation between the initial CT-based diagnostic impression and final stroke subtype classification, even among trained vascular neurologists. Sensitivity and positive predictive value were low for all stroke subtypes; the exception was a high sensitivity for small-vessel disease. The diagnosis of small-vessel disease was overestimated; more than one-third of patients initially considered to have small-vessel disease subsequently had a different final diagnosis.

Significant controversy exists regarding the diagnostic utility of echocardiography in the evaluation of acute ischaemic stroke. Yu and colleagues suggest that its low yield, coupled with lack of evidence for urgent anticoagulation, render it a poor screening tool for stroke patients.^[2] They cite a yield of 3.8% for transoesophageal echocardiography and a yield of 1.5% for transthoracic echocardiography. Other authors believe that echocardiography, especially via the transoesophageal route, strongly influences secondary prevention strategies.^[6–8] An MRI-based diagnostic approach may allow doctors to selectively obtain transoesophageal echocardiography examinations in the subgroup of patients who are most likely to benefit.

In patients with a change in diagnosis, MRA was the key determinant of change in 16%. Sixty per cent of these changes were based on cervical MRA findings and 40% were based on the results of the intracranial MRA. In the subgroup analysis of cervical MRAs, there was excellent agreement between tof-MRA, ce-MRA and ultrasound in patients with minimal atheromatous disease. However, there was considerably less agreement when disease was present. Our findings corroborate previously published results regarding the sensitivity and specificity of MRA in the identification of cervical and intracranial atherosclerosis.^[9–13] We found that MRA had a high negative predictive value and could obviate the need for further vascular imaging if the study was normal; this applied to about 90% of our patients. By contrast, the poor agreement with carotid ultrasound in patients with a positive MRA supports the need for additional diagnostic studies (such as CT angiography or ultrasound).

TIAs were more commonly classified by MRI into the other group (65%) compared with thromboembolic (35.2%) and the small-vessel disease (27.3%) groups (p≤0.001). Most of the TIA patients had negative DWI scans and, therefore, our diagnostic algorithm may not be ideally suited for the diagnostic work-up of TIA patients. The addition of perfusion-weighted MRI (PWI) may detect a lesion in one-third of the DWI-negative TIA patients as suggested by Mlynash et al, although there is insufficient evidence to support its routine use.^{[14 15]} Mlynash found that the combination of DWI and PWI, when performed within 48 h of symptom onset, identified evidence of brain ischaemia in half of the patients admitted with stroke symptoms lasting less than 24 h. Furthermore, long-term Holter monitoring may be more revealing than echocardiography in this subset of patients. A recent study of 24 h Holter monitoring for patients with stroke or TIA found that of the patients with documented paroxysmal atrial fibrillation, 41% presented with TIA.^[16]

This study has some limitations. Our use of the modified TOAST classification at the time of the initial assessment differs from the traditional use of the TOAST classification, which is based on an assessment of all available data from the complete diagnostic evaluation. However, in order to determine the additive benefit of each diagnostic test, pretest assessment was required; we chose a modification of the TOAST classification for its ease of comparison with the final TOAST determination. Moreover, we applied TOAST criteria to patients admitted with TIA; this was also not part of the standard TOAST criteria. However, studies of DWI have demonstrated that many ischaemic episodes with symptoms lasting <24 h are also associated with new infarction. One-third of individuals with clinically defined TIAs exhibit a lesion on DWI.^[17]

These data represent a clinical approach to the determination of stroke aetiology, and not the true ‘yield’ of diagnostic testing. This highlights a key difference between our study and many other investigations assessing the diagnostic utility of specific tests. We assessed which tests led to a change in the diagnosis, that is, the principal determinant of change. We feel that this may be a more clinically relevant approach to determine the optimal evaluation for acute ischaemic stroke patients, as opposed to the traditional notion of diagnostic yield. The diagnostic yield of a study is defined as how often a test is positive, whether or not it has a bearing on the final diagnosis. For example, although moderate and high-risk lesions were noted by echocardiography in 53/241 (22.0%) patients, physicians chose transoesophageal echocardiography as the principal determinant of diagnostic change in only 13/241 (5.4%) patients. The concept of the principal determinant of change was chosen to be a more relevant notion in the clinical work-up.

In conclusion, DWI patterns appear to predict stroke aetiologies better than conventional diagnostic methods. Early MRI/MRA can help tailor the stroke diagnostic evaluation and may limit the need for other studies including transoesophageal echocardiography and carotid ultrasound thereby providing a more cost-effective approach to stroke diagnosis, improved patient satisfaction and decreased length of hospital stay. This algorithm may not be applicable to MRI-negative TIAs as it is based on DWI patterns; the addition of PWI may increase diagnostic sensitivity. Future studies are required to validate this algorithm.