Introduction

Patients with embolic stroke of uncertain source (ESUS) are at high risk for recurrence. The high recurrent stroke risk has been hypothesised to be related to persisting sources of embolisation, which may optimally be prevented with anticoagulation. However, the recently completed ESUS trials demonstrated that neither rivaroxaban1 nor dabigatran2 reduced the rate of recurrent events relative to aspirin. In both trials, patients were selected based on an embolic pattern of infarction and only after a completed workup aimed at identification of embolic sources with established secondary prevention treatments, including anticoagulation and revascularisation. The failure of the two ESUS trials to demonstrate a benefit of anticoagulation has led to re-examination of the ESUS concept. It has been recognised for some time that patients with lacunar pattern infarcts can also have risk factors for embolic stroke.3

Numerous classification systems have been developed to improve the determination of infarct mechanism and ultimately tailored secondary prevention treatments. These include the Trial of Org 10172 in Acute Stroke Treatment (TOAST),4 the refined stop stroke study TOAST (SSS-TOAST),5 as well as the associated causative classification system (CCS),6 and most recently the ESUS criteria.7 While diagnostic criteria vary between these systems, infarcts are ultimately classified as embolic with/without an identified source, or non-embolic, that is, isolated small volume subcortical or lacunar. The latter is best defined using imaging criteria, specifically as subcortical hemispheric or brainstem lesions, with maximum diameters that vary from 15 mm4 8 to 20 mm.5 7 9

In the dabigatran treatment of acute stroke II (DATAS II) trial,10 patients were randomised to dabigatran or aspirin within 72 hours of onset, irrespective of infarct pattern and prior to complete stroke risk factor workup. All patients underwent pretreatment and post-treatment MRI, permitting an in-depth analysis of the relationship between infarct pattern, specific stroke mechanism risk factors and recurrent infarction. In this secondary analysis, we aimed to determine the impact of MRI-defined infarct pattern on aetiological classification and the frequency of recurrent infarction in patients with lacunar versus embolic infarcts.

MethodsPatients

This was a secondary analysis of the DATAS II trial–a phase II, prospective, randomised open-label and blinded end point trial (ClinicalTrials.gov NCT02295826). The data were collected from 2015 to 2018 at six Canadian stroke centres. A total of 305 patients with non-cardioembolic minor stroke or transient ischaemic attack (TIA) (National Institutes of Health Stroke Scale (NIHSS) score≤9, infarct volume≤25 mL) were recruited and randomised to dabigatran or aspirin within 72 hours of onset. Patients with a clear indication for anticoagulation were excluded (as well as treatment with a thrombolytic/endovascular thrombectomy, planned carotid endarterectomy/stent within 30 days, creatinine clearance rate<30 mL/min, ongoing bleeding risks, history of spontaneous intracranial bleeding, contraindication to aspirin or dabigatran or life-expectancy<90 days). MRI was performed before randomisation and repeated on day 30 (with infarct volume measured and infarct location recorded). The primary endpoint was symptomatic haemorrhagic transformation within 37 days of randomisation. Image Analysis. Individual patient MRI data were assessed using standard planimetric techniques (Analyze, V 12.0, Analyze Direct, Overland Park, USA).11 Infarct volumes were measured on diffusion-weighted images (DWI) using standard planimetric techniques. The greatest axial diameter of solitary subcortical or brainstem lesions was measured in order to classify infarcts as isolated small subcortical infarcts (solitary subcortical or brainstem lesion<2 cm on DWI) or embolic, as previously described.5–7

Clinical stroke mechanism classification

Stroke mechanism was adjudicated separately using the TOAST criteria, and the consensus definition of ESUS by an investigator (EC) blinded to the treatment group.4 Stroke mechanism data were obtained from the results of all investigations recorded at baseline and throughout the 90-day study period, including those related to recurrent stroke events. Patients with ≥50% stenosis of an extracranial or intracranial artery supplying the territory of the baseline infarct were classified as ‘large-artery atherosclerosis’. Patients with at least one known high (bioprosthetic cardiac valve) or medium (patent foramen ovale, recent myocardial infarction and congestive heart failure) risk factor for cardioembolism were classified as ‘cardioembolism’. Patients that did not fall into the ‘large-artery atherosclerosis’ or ‘cardioembolism’ categories, without cortical symptoms on NIHSS item scores (item 3—visual fields, item 9—best language/aphasia and item 11—extinction and inattention), and an isolated small subcortical infarct pattern on DWI were classified as ‘small-vessel occlusion’.

DWI-TOAST

Stroke mechanism was reassessed based on the DWI infarct topography (DWI-TOAST). The imaging criteria for isolated small subcortical infarction were modified from the original 15 mm maximum diameter to 20 mm on DWI, as previously described.7 9 Patients without a DWI lesion were classified as ‘TIA’. The remaining patients were classified as ‘unknown’.

Patients were classified as ESUS only if they had embolic pattern infarction on DWI and no identifiable embolic source. Patients with insufficient investigation to satisfy ESUS criteria (ie, lacking echocardiography or both intracranial and extracranial artery imaging) were not classified as ESUS, as described previously.7 Patients with ‘major-risk’ cardioembolic sources were also not classified as ESUS.

Statistical analysis

Statistical analysis was performed using Python, with the NumPy, pandas, scikit-learn, statsmodels, researchpy and SciPy libraries. A one-way analysis of variance (ANOVA) was used to compare infarct pattern and baseline patient characteristics of age and NIHSS. Infarct patterns (embolic and isolated small subcortical infarcts) and infarct volumes in embolic and isolated small subcortical infarcts were compared using a t-test. Pearson’s χ2 tests were used to compare the frequency of embolic (and non-embolic) infarct patterns and recurrence, as well as ESUS (and non-ESUS) patients. A multivariable logistic regression model was used to determine the risk of recurrent stroke according to the baseline DWI-TOAST category.

Patient and public involvement

It was not appropriate or possible to involve patients or the public in the design, conduct, reporting or dissemination plans of our research.

ResultsInfarct pattern classification

A total of 305 patients had infarct patterns classified at baseline. Out of total, 148 patients had infarcts classified as ‘embolic’, and 93 met the criteria for ‘isolated small subcortical infarcts’ on imaging (table 1). In the remaining 64 patients, no infarct on DWI was present, consistent with a TIA. Of the 93 isolated small subcortical infarct patients, 19 (20.43%) had an infarct diameter (on DWI only) between 15 and 20 mm. Another 10 patients with isolated subcortical lesions were classified as embolic due to infarct diameters that were >20 mm. In these 10 patients, the diameter ranged from 20.2 to 27.1 mm.

Table 1

Infarct pattern characteristics

Stroke mechanism classification (TOAST)

The majority of DATAS II patients had stroke/TIA of undetermined cause using TOAST criteria (table 2). The DATAS II exclusion criteria made most patients with cardioembolic stroke ineligible. Thirteen patients had a patent foramen ovale (PFO) or intrapulmonary shunt, two had known congestive cardiac failure, one patient had a recent (<1 year) myocardial infarct, and one patient had a bioprosthetic heart valve. Although 19 and 17 patients had extracranial and intracranial stenoses of at least 50%, respectively, only 15 of these vascular lesions (8 extracranial and 7 intracranial) were in an artery supplying the vascular territory of the DWI lesion. No patients were classified as ‘undetermined due to two or more possible identified causes of stroke’. Of the 64 patients without a DWI lesion (TIA), two had potential cardiac sources of embolism. Two TIA patients with cortical symptoms were classified as ‘Stroke of Undetermined Etiology’. The remaining 60 were classified as ‘Small-vessel Occlusion’.

Figure 3

Baseline diffusion-weighted imaging lesions of all patients with newly diagnosed atrial fibrillation/flutter (Patient 6 had no lesion). Patient 3 and patient 6 were classified as ‘small-vessel occlusion’ by both Trial of Org 10172 in Acute Stroke Treatment (TOAST) and diffusion-weighted imaging-TOAST (DWI-TOAST) criteria.

Effect of DWI on stroke mechanism classification

After DWI analysis, 19 (6.2%) patients originally classified as cryptogenic stroke by TOAST criteria, were re-classified as ‘small-vessel occlusion’ based on solitary subcortical infarct topography (DWI-TOAST, table 2). Although most patients had congruous clinical and MRI findings, 5/88 (5.7%) patients with isolated subcortical infarcts had cortical signs on examination (figure 1). Four of these patients had lesions in or adjacent to the thalamus presenting with inattention, visual field deficits or aphasia. The fifth patient had an infarct in the left corona radiata adjacent to the splenium of the corpus callosum resulting in a visual field deficit. These five patients were classified as ‘Stroke of Undetermined Etiology’ on TOAST and ‘Small-vessel Occlusion’ on DWI-TOAST.

Figure 1

Baseline diffusion-weighted imaging (DWI) lesions of all patients with isolated subcortical infarcts and ‘cortical’ symptoms on National Institutes of Health Stroke Scale (NIHSS). Patient 1 demonstrated inattention. Patients 2 and 3 demonstrated partial hemianopia and patients 4 and 5 demonstrated aphasia.

Table 2

Trial of Org 10172 in Acute Stroke Treatment (TOAST) and diffusion-weighted imaging-TOAST (DWI-TOAST) classification and baseline clinical risk factors

Embolic stroke of undetermined source

78 patients (25.57% of the study population) met the criteria for ESUS. 15 patients did not meet ESUS criteria, due to ≥50% extracranial/intracranial stenosis in an artery supplying the area of the DWI lesion. Due to the requirement for echocardiography, which was not standard at all trial sites or part of the DATAS II trial protocol, 118 patients could not be classified using ESUS criteria. Another 60 patients did not have either intracranial or extracranial vascular imaging, which also precluded ESUS classification.

Ischaemic infarct recurrence

New lesions were detected in 23 patients on the day 30 repeat MRI scan. Recurrent infarcts were visible on DWI (22/23) or T2 weighted images (1/23). Most recurrent lesions (17/23, 73.9%) were covert, that is, not clinically identified as stroke. Another four patients had recurrent stroke symptoms with no lesion on imaging (TIA).

DWI-TOAST subtype and recurrence

Of the 27 patients with recurrent infarcts, 16 occurred in patients with undetermined aetiology at baseline using DWI-TOAST criteria. The relative risk of recurrent infarction in patients with undetermined aetiology was increased (standardised coefficient=1.0 (0.1, 1.9), p=0.029; table 3). While the majority of recurrent infarcts in patients with undetermined aetiology at baseline also had an embolic pattern, 5 (31.3%) developed isolated small subcortical infarcts on day 30 DWI (figure 2).

Table 3

Multivariate logistic regression model of recurrent stroke risk according to baseline diffusion-weighted imaging-Trial of Org 10172 in Acute Stroke Treatment (DWI-TOAST) categories

Figure 2

(A) Sankey diagram demonstrating baseline and recurrent infarct patterns in patients with a change in pattern between baseline and day 30. (B) Axial diffusion-weighted imaging (DWI) of all five patients with embolic pattern infarcts at baseline and isolated small subcortical pattern infarcts on day 30 (bottom).

Atrial fibrillation incidence during the trial

Six patients were diagnosed with paroxysmal atrial fibrillation/flutter (AF) between baseline (randomisation) and day 30, all of whom were anticoagulated and did not have a recurrent event. Three of the six patients were classified as ‘stroke of undetermined etiology’ at baseline (figure 3). One of the AF patients was classified as ‘cardioembolic’ at baseline due to a patent foramen ovale. Two patients with AF were classified as ‘Small-vessel Occlusion’ at baseline, including one TIA and one patient with a 12.5 mm diameter peri-thalamic internal capsule lesion on baseline DWI.

Discussion

In this retrospective analysis, definitive identification of infarct topography in patients with minor ischaemic stroke/TIA resulted in a change in aetiological subtype in 24 (7.9%) of cases. Most commonly, this change was from cryptogenic to small-vessel occlusion, but the converse, that is, cortical signs with isolated subcortical infarcts was also seen. A total of 40.3% of patients in DATAS II had embolic pattern infarcts of unknown aetiology (cryptogenic), which is comparable to that seen in prior investigations of stroke aetiology (17% to 39%).7 This group was associated with a high rate of recurrent events (13.0%). Although recurrent infarcts were most often also topographically embolic, isolated small subcortical infarcts also occurred in 4/16 (25%) patients with a cryptogenic baseline event. The variable relationships between MRI-defined infarcts and both baseline clinical signs, as well as recurrent infarction patterns, is a challenge to both the lacunar and ESUS hypotheses.3 7

Despite significant improvements in investigations aimed at optimising secondary prevention, the determination of mechanism remains a challenge. Our results indicate that in the absence of DWI, aetiological classification based on clinical grounds, results in more patients being classified as cryptogenic, due to poor identification of isolated small subcortical infarcts on CT. DWI-TOAST classification resulted in the re-classification of many of these patients to small-vessel occlusion. In patients with small embolic pattern infarcts seen on DWI, but not visible on CT, classification was unchanged. While patients with minor ischaemic stroke/TIA have previously been shown to have an elevated recurrent stroke risk,12 13 the present findings suggest baseline topography is also an important factor. Indeed, these patients were those specifically targeted in the ESUS trials.1 2 7 The fact that recurrent infarcts in these patients were not always embolic (figure 2) may be related to the failure of these trials to demonstrate a benefit with anticoagulation, although a failure of large-vessel atheromatous changes to respond to the latter have also been hypothesised to be relevant.14

DWI is the effective gold standard for the definitive identification of infarct pattern/location. It has previously been demonstrated that relying on clinical features and non-contrast CT alone results in a high rate of TOAST misclassification.15 Even when this is performed routinely, as in the present study, the actual mechanism of a stroke can only be inferred. The recurrent events seen in patients with embolic pattern cryptogenic strokes in the present study illustrate the limitations of this approach (figure 2). In the patients with recurrent isolated small subcortical infarcts, it is unknown if this represents new micro-emboli to small perforating arterioles, or a completely independent mechanism, that is, small-vessel occlusion. While the close temporal relationship between the infarcts may suggest a common aetiology, the common risk factors for both embolic and small-vessel occlusion strokes cannot be discounted.16–18

The relationship between infarct topography and mechanism is further complicated by arbitrary definitions of small subcortical infarcts. The original TOAST criteria formulated in the pre-DWI era limited this subtype to infarcts <15 mm in diameter. Modifications to the TOAST criteria extended the limit to 20 mm on DWI, as was used in the present study.19–26 While this modified definition added 19 patients (6.2% of the trial population) to the small-vessel occlusion group, there were another 10 patients with isolated subcortical infarcts between 20 and 30 mm in diameter. Increasing diameter makes an arteriolar infarct less likely, but the true mechanism of occlusion in these isolated infarct cases is impossible to determine.

In the present analysis, stroke recurrence rates were lowest in the small-vessel occlusion group, which is consistent with the findings of a 2004 meta-analysis.27 In addition, the recurrence rate in the large-artery atherosclerosis group of 13.3% at day 30 is comparable to the 9.2% recurrence rate by 3 months in the Oxfordshire Community Stroke Project and a more recent analysis by Amarenco et al 27 28

The proportion of patients with ESUS in the present study (25%) is comparable to previous studies,29 30 but lower than expected, given the proportion of patients with embolic topography, meeting TOAST cryptogenic criteria. This reflects one of the key weaknesses of the study. A high proportion of patients did not meet ESUS consensus (54/305; 17.7%) criteria due to insufficient investigation. The DATAS II trial was conducted during a time of evolving investigation for stroke. While intracranial and extracranial CT angiography are now standard in most centres, this was not the case during the study period. However, even now there is considerable variability with respect to echocardiographic and extended cardiac rhythm monitoring after stroke.31–34 Irrespective of whether patients in our study met the criteria for ESUS, the recurrence rate in patients with embolic topography and a cryptogenic DWI-TOAST subtype had the highest recurrent event rate, consistent with prior studies.35

The demonstration that patients presenting with cortical clinical syndromes can sometimes have isolated subcortical infarcts further complicates the inferential relationship between infarction pattern and mechanism. It has previously been shown that patients with subcortical syndromes often have small cortically based infarcts on MRI,15 but we found the opposite in the present study. This likely reflects proximity to the thalamus in most cases but given the transient and minor stroke population included in our study this may also reflect transient cortical ischaemia, with infarction seen only in the subcortical regions of the territory initially affected.

Our findings suggest that cryptogenic embolic pattern infarcts should be considered the highest risk for recurrent events. In the absence of DWI, however, this group of patients cannot be accurately identified. Additional targeted treatment trials in the embolic infarct pattern population are required to determine if the high recurrent rate can be reduced. In the interim, all stroke patients require extensive investigations for all potential embolic sources.