Undetermined stroke with an embolic pattern--a common phenotype with high early recurrence risk.

Abstract Introduction. Undetermined strokes with an embolic pattern (USEP) represent a common phenotype. We assessed their frequency and compared USEP with cardioembolic stroke with a known source and non-cardioembolic stroke etiology. Methods. Study patients were 540 consecutive ischemic stroke patients admitted to Helsinki University Hospital with primary end-point of recurrent stroke in a 21-month follow-up. Cox regression adjusting for CHA2DS2-VASc and anticoagulation estimated the risk of USEP on recurrent stroke. Results. A total of 229 (42.4%) patients had a non-cardioembolic stroke etiology, 184 (34.1%) had a cardioembolic stroke with a known source, and 127 (23.5%) were classified as USEP. USEP patients had less diabetes and prior TIA, with more severe symptoms than the non-cardioembolic stroke cases. They were younger, had fewer comorbidities, and less severe symptoms than the cardioembolic stroke patients. Cumulative risk of recurrent stroke was 10.0% (95% CI 4.1%–15.9%) for USEP, 5.0% (1.1%–8.9%) for cardioembolic strokes, and 5.0% (3.0%–7.0%) for non- cardioembolic strokes (P = 0.089). USEP associated with a higher risk of recurrent stroke compared to non-cardioembolic strokes (hazard ratio 2.36, 95% CI 1.02–5.47; P = 0.046) and cardioembolic stroke with a known source (1.83, 1.07–3.14; P = 0.028). Conclusions. Despite their younger age and more favorable risk factor profile compared with other phenotypes, USEP exhibited a high risk of stroke recurrence.


Introduction
Of all ischemic strokes, etiology in 12% to 39% is classifi ed as undetermined -i.e. a plausible causative mechanism for the stroke is not identifi ed (1). According to the literature, 60% to 87% of such strokes are non-lacunar, with embolic neuroimaging trials ongoing. However, thus far only one study has systematically assessed the frequency of ESUS, fi nding a 10.0% prevalence, while 13.4% remained undetermined other than ESUS, mostly due to incomplete diagnostic work-up (18). In clinical routine, full diagnostic examinations are not carried out in all consecutive patients due to e.g. patient non-compliance, limitation of care due to poor prognosis, or limited resources. Prognostic data on strokes of undetermined cause with embolic features are scarce, and comparisons of ESUS patients with other undetermined stroke patients have not been reported.
We thus pragmatically assessed the frequency, comorbidities, and risks of recurrent stroke and death in consecutive patients with undetermined stroke with an embolic pattern (USEP) sharing embolic lesion topography in neuroimaging irrespective of completeness of diagnostic evaluation. We compared established cardioembolic and non-cardioembolic strokes with USEP. Within USEP, we compared ESUS cases with those not meeting the ESUS criteria.

Methods
Th is retrospective observational cohort study was carried out at the Helsinki University Hospital and approved by the relevant institutional authorities. Informed consent was not required nor sought as the study setting was observational without direct patient contact. Our hospital serves as the only university teaching hospital in the province of Uusimaa, with a catchment population of 1.6 million. Th e hospital has the only neurological emergency room with 24/7 service in the province.
We identifi ed all consecutive patients from the hospital ' s electronic discharge database from 1 January to 31 March 2010 and 1 January to 31 March 2012 with diagnosis of I63 according to 10th version of the International Classifi cation of Diseases and resident in Finland. Th e patients were compiled into a single data set with uniform follow-up of 21 months (or until death) for all cases.
Subsequently, all electronic patient records, including ECGs, laboratory, radiological, and medication prescription database were retrieved for review from a province-wide electronic patient data archive. Pre-existing comorbidities recorded included hypertension (existing diagnosis or any blood pressure medication), diabetes (existing diagnosis or any antidiabetic medication), dyslipidemia (antilipemic medication or total cholesterol Ն 5.0 mmol/L, low-density lipoprotein Ն 3.0 mmol/L, high-density lipoprotein Ͻ 1.0 mmol/L, or triglycerides Ն 2.0 mmol/L), smoking (current, smoking at least Ն 1 cigarette/day within the last year; ex-smoker, quit Ͼ 1 year prior to stroke), prior history of coronary heart disease, congestive heart failure, myocardial infarction (MI), peripheral arterial disease, transient ischemic attack (TIA), stroke, AF, and cancer. Pre-stroke AF was classifi ed as persistent/permanent or paroxysmal. Pre-stroke CHA 2 DS 2 -VASc score was calculated for each patient as follows: congestive heart failure (1 point), hypertension (1 point), age Ն 75 years (2 points), diabetes (1 point), prior stroke or TIA (2 points), vascular disease (1 point), age 65 -74 years (1 point), and female sex (1 point) (19). If female sex was the only risk factor, it was not scored on CHA 2 DS 2 -VASc. Stroke severity on admission was assessed with the National Institutes of Health Stroke Scale (NIHSS). Acute stroke treatments and post-stroke antiplatelet and oral anticoagulation regimen was recorded.
All patients underwent brain scan at least once with CT or MRI, a chest X-ray, and a 12-lead ECG on admission and at least once during hospitalization. Imaging of extracranial cervical arteries was performed as judged by the treating physician either with Doppler ultrasound, CT angiography, or MR angiography, or combinations of these. Imaging of intracranial arteries was done with CT angiography or MR angiography. In patients with sinus rhythm and no history of paroxysmal AF, AF was screened either with: 1) continuous ECG monitoring during their stay in the emergency room, stroke unit, or intensive care unit; 2) repeated ECG; and/or 3) 48-hour Holter ECG. Th e main echocardiography parameters were recorded: left atrium diameter, left ventricle end-diastolic diameter, and left ventricle ejection fraction. Generally, the Finnish publicly funded health care system encourages performing ancillary diagnostic investigations only if their results potentially aff ect treatment decisions.
Board-certifi ed neuroradiologists initially evaluated all brain and artery imaging scans. Lesions were defi ned as embolic if solitary or multiple superfi cial, cortical or cerebellar, large deep, or superfi cial and deep combined, in the absence of explanatory small-vessel disease or atherosclerotic stenosis in a relevant artery, or other uncommon cause such as dissection, vasospasm, vasculitis, or hemodynamic compromise. In case no embolic pattern was visible in baseline or follow-up brain imaging, embolism was supported by rapid recanalization of an occluded main brain-supplying artery or its branches or a computed tomography perfusion lesion compatible with embolic pattern. Stroke etiology was fi rst classifi ed according to Trial of Org in Acute Stroke Treatment (TOAST) (20). Subsequently, according to clinicoradiological characteristics we classifi ed patients as (A) non-cardioembolic stroke etiology (TOAST subgroups: 1, large-artery atherosclerosis; 3, small-vessel disease; 4, other determined etiology; and 5a, multiple concomitant causes), (B) cardioembolic stroke with a known high-risk source (TOAST subgroup 2, i.e. probable cardioembolism), and (C) USEP (TOAST subgroups 5b and 5c). Imaging-negative cases with undetermined etiology did not qualify for USEP and thus fell into the non-cardioembolism etiology group. High-risk sources for cardioembolism were those with arbitrary Ն 2% primary risk for ischemic stroke, as previously described (21), including AF or fl utter, intracardiac thrombus, prosthetic cardiac valve, myxoma or other cardiac tumors, mitral stenosis, recent ( Ͻ 4 weeks) MI, left ventricular ejection fraction Ͻ 30%, valvular vegetations, or infective endocarditis. All uncertain or low-risk sources for cardio embolism, such as patent foramen ovale, atrial septal aneurysm, or mitral valve calcifi cation or prolapse, were classifi ed as USEP in the presence of embolic lesion(s) in neuroimaging.
Among the USEP group, cases fulfi lling the ESUS criteria according to NAVIGATE ESUS trial (https://clinicaltrials.gov/ ct2/show/NCT02313909, accessed 7 January 2015) were identifi ed as meeting the following criteria: non-lacunar stroke, absence of cervical carotid atherosclerotic stenosis Ն 50% or occlusion, no AF in history or aft er Ն 24-hour cardiac rhythm monitoring, no intracardiac thrombus on transthoracic echocardiography, and no other specifi c cause identifi ed.
Since stroke patients are invariably hospitalized and treated in public hospitals in our country, we performed follow-up by reviewing all medical records in the province-wide electronic patient data archive. Mortality data came from Statistics Finland. Follow-up time was 21 months or until death. Th e primary endpoint was non-fatal or fatal recurrent stroke. Secondary endpoints were death from any cause and a composite of recurrent stroke or death, whichever occurred fi rst.
Median and interquartile range (IQR) are reported for all continuous variables with non-normal distribution. Groups were compared with chi-square, Fisher ' s exact, Mann -Whitney U , and ANOVA tests where appropriate. Kaplan -Meier curves were plotted for end-points and groups defi ned by phenotypic  classifi cation compared with the log rank test. Life tables were used to calculate cumulative risks and their 95% confi dence intervals (CI) of the end-points. Cox proportional hazards models were constructed to assess the impact of USEP on the end-points. All models were adjusted for pre-stroke CHA 2 DS 2 -VASc score and post-stroke anticoagulation use. CHA 2 DS 2 -VASc score was used as an ordinal variable. Models including mortality end-point were also adjusted for baseline NIHSS score as an ordinal variable. Since CHA 2 DS 2 -VASc score integrates the key parameters relevant to the studied end-points and has been shown to correlate accurately with the risk of stroke or death also in non-AF patients (22), and to avoid over-adjustment, no further correction for comorbidities was made. Two-sided statistical signifi cance was set at 0.05. All analyses used IBM SPSS 20 (IBM Corp, Armonk, NY, USA).
Compared with non-cardioembolic strokes, USEP patients had less oft en diabetes mellitus (22% versus 13%) and prior TIA (8% versus 2%), but more severe stroke symptoms (median NIHSS Data are median (interquartile range), or n (%). CT ϭ computed tomography; ER ϭ emergency room; MI ϭ myocardial infarction; MRI ϭ magnetic resonance imaging; NIHSS ϭ National Institutes of Health Stroke Scale. a P Ͻ 0.05 for undetermined stroke with an embolic pattern versus non-cardioembolic stroke etiology. b P Ͻ 0.05 for undetermined stroke with an embolic pattern versus cardioembolic stroke with a known source. c Transesophageal echocardiography in 2, 1, and 5 patients, respectively. d P Ͻ 0.01 for cardioembolic stroke with a known source versus each of the other two groups.
score 2 versus 4). Compared with patients with cardioembolic stroke with a known source, USEP cases were younger (median age 72 versus 66 years), had less frequently heart failure (17% versus 5%) and lower CHA 2 DS 2 -VASc score (median 3 versus 2), and less severe stroke symptoms (median NIHSS score 5.5 versus 4). Antiplatelet medication was introduced in USEP patients as oft en as in non-cardioembolic stroke patients (85%) ( Table I). In transthoracic echocardiography, left ventricular ejection fraction was lower among those with cardioembolic stroke with a known source (55% versus 60%).
Of the USEP patients, 46 (36.2%; 8.5% of all patients) fulfi lled the diagnostic criteria for ESUS. Th ose not fulfi lling the ESUS criteria were older, had more frequently hypertension and cancer, exhibited higher CHA 2 DS 2 -VASc scores, and were less oft en treated with intravenous thrombolysis. Th e main echocardiographic parameters were similar in both groups (Table II).
During the 21-month follow-up, a total of 65 deaths from any cause, 30 non-fatal or fatal recurrent strokes, and 92 composite events of recurrent stroke or death had occurred. Cumulative 21-month risk of recurrent stroke was doubled for USEP (10.0%, Table II. Comparison of patients with undetermined stroke with an embolic pattern meeting the diagnostic criteria for embolic stroke of undetermined source (ESUS) ( ϩ ), to those not meeting the criteria, ESUS ( -  Figure 3A; log rank P ϭ 0.089). Compared with the group of non-embolic stroke, the 21-month cumulative risk of death was higher in the high-risk cardioembolism source group (7.0%, 95% CI 3.1% -10.9% versus 18.0%, 12.1% -23.9%), log rank P ϭ 0.004 for overall comparison. Cumulative risk of death was 12.0% (6.1% -17.9%) for USEP ( Figure 3B). Cumulative 21-month risk of the composite end-point of stroke or death was 11.0% (7.1% -14.9%) for non-embolic stroke, 22.0% (16.1% -27.9%) for cardioembolic stroke with a known source group, and 20.0% (12.2% -27.8%) for USEP, being thus signifi cantly higher ( P ϭ 0.008) for the two last-mentioned compared with non-embolic stroke ( Figure 3C).
Cumulative rates of the end-points at 12 months appear in Supplementary Table I   In the Cox regression adjusted for CHA 2 DS 2 -VASc score and anticoagulation use, USEP independently was associated with higher risk of recurrent stroke as compared with those with non-cardioembolic stroke (hazard ratio 2.36, 95% CI 1.02 -5.47; P ϭ 0.046). Compared with cardioembolic stroke with a known source, USEP was also associated with higher risk of recurrent stroke (1.83, 1.07 -3.14; P ϭ 0.028). Regarding secondary endpoints, there were no statistical diff erences between USEP and non-cardioembolic stroke in the risks of death from any cause or composite of recurrent stroke or death. However, compared with cardioembolic stroke, risk of death was lower for the USEP group (0.33; 1.17 -0.65; P ϭ 0.001), with a similar trend for the composite end-point of recurrent stroke or death (0.60, 0.34 -1.07; P ϭ 0.083).

Discussion
Th e main result of our study was that non-lacunar strokes with an embolic imaging pattern without an identifi ed cause accounted for about one-quarter of all the consecutive ischemic strokes and three-quarters of the undetermined strokes. Considering the various characterizations based on the lesion size and/or location, this proportion is in accordance with prior reports (60% -87%) (2 -4). In our study, USEP subtype was independently associated with a higher risk of recurrent stroke already in a 21-month follow-up in comparison to those with non-cardioembolic stroke or cardioembolic stroke with a known source.
Our proportion of ESUS, 8.5%, is in accordance with the recent Athens Stroke Registry data (10.0%) (18) despite our underutilization of continuous ECG monitoring. Th e NAVIGATE ESUS trial (2) applied modifi ed ESUS criteria, which we also used, and which do not necessitate intracranial artery imaging. In case of using the original diagnostic criteria (2), the proportion of our ESUS patients falls to 5.4% mainly due to the under-utilization of intracranial artery imaging. Th e frequency of potential ESUS patients eligible for a prospective trial among all the USEP cases could thus be much higher than 1/10 with a more systematic screening.
With regard to comorbidities, notable diff erences emerged between USEP and patients with embolic stroke with known source. USEP patients were younger with less cardiovascular comorbidity using the CHA 2 DS 2 -VASc score. Compared with those with non-embolic stroke, USEP patients had less oft en diabetes mellitus and prior TIA. Stroke symptoms were more severe in USEP compared with non-cardioembolic stroke but less severe compared with cardioembolic stroke with a known source. Th ese observations may at least partly point out that short paroxysms of AF may frequently underlie USEP, since strokes in the setting of persistent or permanent AF have been more severe compared with strokes due to paroxysmal AF (23). Notably, the antithrombotic medication scheme was identical in non-cardioembolic strokes and USEP patients.
Among the USEP patients baseline diff erences emerged between those fulfi lling the ESUS diagnostic criteria and those who did not. Since the cases not fulfi lling the criteria were older with more comorbidities, the clinician may have judged further end-point events seems rather unlikely, as we had access to provincewide electronic medical records, including imaging and laboratory records. Regarding fatal events, the quality of the data maintained by Statistics Finland has been assessed to be good (28).
In conclusion, acute embolic strokes in neuroimaging without established source were common among consecutive ischemic stroke patients and associated with a high early risk of recurrent stroke. Th e number of patients fulfi lling the ESUS criteria can be increased by a more thorough case screening by utilizing existing diagnostic methods, and our fi ndings add to the rationale for testing whether anticoagulants are superior to antiplatelets in the ESUS patients. However, developing novel methodse.g. continuous ECG monitoring devices, cardiac imaging, screening for thrombophilia, imaging-screening or biomarkers for malignancies -to discover the potential causes of stroke seems also justifi ed (27,29). Finally, further work is needed to standardize the entity of undetermined stroke and establish prognosis in its subgroups.
investigations unnecessary. As stroke severity was equivocal in both USEP subgroups, it is unlikely that this factor would have infl uenced the decision whether to carry out complete examinations. However, data on the specifi c reasons why complete investigations were not carried out by the individual clinician were not available.
Despite the more favorable clinical profi le at baseline, USEP classifi cation independently was associated with recurrent stroke in a relatively short follow-up of only 21 months. Our fi ndings of the risk of recurrent stroke among the USEP cases (7.0% at 12 months) is within the limits of most earlier reports in patients with undetermined stroke, with annual rates ranging from 2.5% to 7.8% (3,4,12 -15). However, a Korean study described a 30% recurrence rate at 12 months for such patients (24). Compared to our rate (16.0% at 12 months), a combined end-point of recurrent stroke or death occurred at a markedly lower rate (5.8% per year) in the subgroup of cryptogenic strokes in the prospective WARSS trial comparing aspirin with warfarin in secondary stroke prevention (25). Th e relatively low number of end-points in our series precludes making fi rm conclusions regarding the risks stratifi ed by the USEP subgroups. However, frequency of recurrent strokes appeared similar in the patients fulfi lling the ESUS criteria and those not fulfi lling them (Supplementary Figure 1A to be found online at http://informahealthcare.com/doi/abs/10.3109/07853890. 2015.1057612).
High early mortality among our USEP patients not meeting the ESUS criteria may refl ect their expected poor outcome and withdrawal of complete examinations (Supplementary Figure  1B to be found online at http://informahealthcare.com/doi/abs/ 10.3109/07853890.2015.1057612). Existing data on mortality in undetermined stroke are variable and sparse. Th e few reports showed 16.0% mortality at 3 months (26), 17% at 6 months, 3.3% at 1 year (3), and 8.0% at 2.1 years (12). Notably, a populationbased study conducted in the 1980s showed a 48.6% mortality rate for ischemic stroke of uncertain cause (14). Th e rate variation probably refl ects the varying criteria for defi ning undetermined or cryptogenic stroke as well as population and study period diff erences. Our fi ndings suggest an approximately 10% mortality at 1 year in USEP patients, the rate being, however, considerably smaller in those fulfi lling the ESUS criteria.
Based on the recent findings on AF detection trials, a large proportion of patients with cryptogenic stroke may have occult atrial fibrillation (16,17). In the CRYSTAL-AF trial, an implantable loop recorder detected annually ∼ 10% of new AF in a 36-month follow-up. Since our pragmatic USEP defi nition covered also patients with incomplete diagnostic evaluationrather characterizing the neuroimaging presentation than underlying pathophysiology -it likely would capture some cases with a high-risk source for cardioembolism. Furthermore, e.g. intracranial atherosclerosis and subclinical malignancies may have gone unnoticed when not screened systematically (27).
Our study has strengths and limitations. Th e main strengths are that it describes the phenotypic presentation in a real-life consecutive patient population without exclusions and documents the diagnostic work-up and follow-up in detail. Th e main limitations include its single-center setting and relatively low number of end-point events. Some selection and referral bias is also possible, since our center is a tertiary teaching hospital facility, and some non-acute and previously non-independent cases are treated in regional hospitals. Although we had access to province-wide comprehensive electronic patient data archives and all patients with stroke are generally hospitalized in our health care system, we might have missed a few events, for instance if treated outside the hospital catchment area or if neurological symptoms have gone unnoticed in severely disabled patients. Inappropriate ascertainment of