Contribution of routine cardiac biological markers to the etiological workup of ischemic stroke

a b s t r a c t

Background: Optimization of the detection of atrial fibrillation following stroke is mandatory. Unfortunately, ac- cess to long-term cardiac monitoring is limited in many centers. The aim of this study was to assess the potential usefulness of three routine biological markers, troponin, D-dimers and BNP, measured in acute stroke phase in the selection of patients at risk of cardio-embolic stroke.

Methods: Troponin, D-Dimers and BNP were measured within 48 h after admission for ischemic stroke in 634 pa- tients. Stroke mechanism was defined at the 3 months follow-up visit using ASCOD classification using a stan- dardized work-up. Association between clinical, radiological and biological markers and stroke mechanism was evaluated using logistic regression analyses.

Results: 159 patients (25.1% of total study population) had a cardiac mechanism. On multivariate analysis, admis- sion initial Stroke severity (OR 1.04, 95 CI% 1.004-1.07, p b 0.05) history of heart failure (OR 3.03, 95% CI 1.19- 7.73, p b 0.05), ECG abnormalities and high BNP value (OR 4.34, 95% CI 2.59-7.29, p b 0.05) were associated with pure cardiac stroke mechanism.

Conclusion: High BNP value measured within 48 h after stroke admission is an independent predictor of cardiac stroke mechanism. Its measurement might be used to improve the selection of patients for whom further cardio- logic investigations such as continuous long term ECG monitoring would be the most useful. BNP should be added to the standard admission-work-up for stroke patients.

(C) 2018


risk of stroke recurrence after a first ischemic event is estimated around 10% in the first year and reaches 20 to 30% at five years [1]. Early recognition of stroke mechanism is a critical step to optimize sec- ondary prevention. Indeed, among the wide sources of ischemic stroke, the identification of cardiac causes is of special interest because it di- rectly leads to the introduction of anticoagulants [2]. Atrial fibrillation accounts for 50% of cardioembolic strokes, but could also be an un- derlying cause of 30% of embolic stroke of unknown sources [3,4]. Therefore the optimization of the detection of AF following stroke is mandatory. It is well known that increased duration of ECG recording is associated with higher rates of AF detection [5]. However the avail- ability of external long term monitoring and implantable devices is still limited and cannot be performed to every stroke patient with un- known stroke mechanism, suggesting the need to identify a set of bio- markers that could improve the selection of patients that would benefit from these investigations.

* Corresponding author at: stroke unit, Bordeaux University Hospital, 33076 Bordeaux, France.

E-mail address: [email protected] (J.S. Liegey).

Currently some heart rhythm characteristics (i.e. frequent prema- ture beats) or cardiac morphologic parameters (i.e. left atrial dilation) are the most well known predictors [6]. In addition, three biomarkers, widely available in the clinical setting, showed promising results in this context: the Brain Natriuretic Peptid (BNP) and NT-proBNP [7] reflecting cardiac dysfunction, troponin [8,9], reflecting cardiac ische- mia, and D-dimers [10], a marker of thrombosis. However most of these studies used advanced biochemical techniques to measure the levels of these biomarkers and did not address their value when mea- sured in the context of a routine evaluation. The aim of this study was investigate the association between a “triple positivity” of this set of bio- markers and the presence of a cardiac source of stroke in order to deter- mine their potential usefulness in the selection of patients requiring long-term cardiac monitoring.

Material and methods

Study population

We performed a retrospective monocentric study by gathering data from electronic medical records of patients who were admitted in our stroke unit from January 2016 to October 2016.

0735-6757/(C) 2018

Patients were eligible if they had had an ischemic stroke confirmed on brain imaging (CT or MRI), were part of the regional stroke observa- tory (ObA2) and had troponin, D-dimers and BNP levels available. Pa- tients were excluded from the study if they had an alternative diagnosis to stroke (including TIA), if they had not benefited from at least 48 h telemetry in our unit, or if one or more biological data were missing among troponin, BNP and D-dimers levels. For every patient, demographical data (sex, age, NIHSS score at admission, pre-stroke and discharge Rankin score) and medical history (AF, diabetes, Arterial hypertension, heart failure, stroke and vascular disease) was collected, as well as the CHA2DS2-VASC score [11], regardless of any history of AF and without taking into account the current stroke. Administration of intravenous thrombolysis or Mechanical thrombectomy was documented.

Clinical protocol

Biomarkers samples were drawn within 48 h after admission (D-di- mers levels had to be measured before intravenous thrombolysis if any). According to our lab, normal values were as follow: BNP b100 pg/mL (chemiluminescent immunoassay), troponin I b0,04 ng/mL (chemilu- minescent immunoassay), D-dimers b500 ng/mL (ELISA). For patients over 50 years old, we calculated the normal limit value of D-dimers ad- justed to age (bage x 10 in ng/mL) [12]. When biomarkers were col- lected at several time points, only the first one was considered in the analysis. For multivariate analysis, a global biological score was established by adding up the number of elevated biomarker levels, ranging from 0 to 3.

Each patient had a brain imaging (MRI or CT) to establish the diagno- sis of stroke and its unique or multiples locations. Etiological workup in- cluded extra- and intracranial assessment of cerebral arteries using CT or MR angiography, ECG, at least 48 h telemetry in the intensive stroke care unit, 24 h-Holter-ECG, TTE and/or transesophageal echography) and laboratory testing including at least complete blood cell count, stan- dard coagulation tests, Serum electrolytes, assessment of liver and renal function, CRP, HbA1c and lipid profile). No internal or external continu- ous long term (N48 h) ECG monitoring was performed for these patients.

Stroke categorization

Stroke etiology categorization according to the ASCOD classification

[13] was established at 3 months during a follow-up visit. If none or more than one of the 5 potential stroke mechanisms from the ASCOD classification were considered as a potential cause, stroke mechanism was considered as undetermined. Patients for which a cardiac pathology was the cause of stroke, thus ranked “C1”, were group 1. This group was compared to group 2, which included patients with another mechanism than cardiac pathology ranked “1” and those with none or more than one of the 5 mechanism in the classification ranked “1”.

Statistical analysis

Results from quantitative variables were presented using averages clinical results”>and standard deviations (SD) for age and NIHSS score, and with me- dians and extreme values for Rankin score and CHA2DS2-VASC score. Qualitative variables were described in frequency and proportions. Comparisons between groups 1 and 2 were done using Student’s t– test or Fisher’s exact test for parametric and non parametric variables. Distribution of variables CHA2DS2-VASC and Rankin score were com- pared between the 2 groups using ANOVA analysis.

We evaluated the clinical and paraclinical variables associated with a cardioembolic stroke mechanism by studying the association between these variables and groups 1 and 2 with a logistic regression model in a univariate analysis. Two models of multivariate analyses were built: model 1 included the main clinical parameters and each of the 3

biomarkers, model 2 included CHA2DS2-VASC and a global biological score for biomarkers ranging from 0 to 3. Variables with a p b 0.2 in the univariate analysis were included in a multivariate analysis. Vari- ables with a p N 0.2 composing the CHA2DS2-VASC were entered in model 2. p b 0.05 was considered statistically significant. All statistical analysis were done using IBM(C) SPSS(C) Statistics Base 22.0.


During the study period, 857 potentially eligible patients were iden- tified. 223 patients were excluded because of incomplete biological data. 634 patients were included, 159 of them (25.1% of total study pop- ulation) were part of Group 1 (C1 mechanism). Fig. 1 represents the pa- tient flow-chart.

Group classification

In group 1, AF was the most common cause (135 patients, 84.9%). AF was known before the ischemic event for 51.8% of the population, and discovered in the stroke unit for 55 patients, either on admission ECG (n = 24), telemetry (n = 30) or in-hospital 24 h-Holter ECG (n = 1 among 10 recordings). Diagnosis of AF was made between discharge from the stroke unit and the 3 months follow-up visit for 10 patients, based on the results of 24 h-Holter ECG.

In group 2, 219 patients (46.3% of group 2) had another mecha- nism than cardioembolism ranked “1” (atherosclerosis, small-vessel disease, dissection, other causes) at the 3 months follow-up visit. For 236 patients (49.7% of group 2), stroke mechanism was considered undetermined either because of several mechanisms ranked “1” (20 patients with AF, 4.2% of group 2) or because none of the mech- anism ranked “1”.

Comparisons between group 1 and 2

Comparisons between group 1 and 2 are represented in Table 1.

Demographical and clinical data

Groups 1 and 2 did not differ significantly in terms of sex ratio, with a male predominance in both groups (respectively 56% and 62.3%). Age was significantly higher in group 1 (75.3 y.o vs. 67.1, p b 0.05). Pre- stroke Rankin score did not differ between the 2 groups. Patients from group 1 had significantly more severe stroke (mean admission NIHSS 7.8 vs. 4.5) and worse outcome at 3 months.

Cardiovascular comorbidity and CHA2DS2-VASC

CHA2DS2-VASC score was significantly higher in group 1, due to an over-representation of 3 items: history of hypertension (67.9% vs. 56.4%, p b 0.05), heart failure (10.1% vs. 2.1%, p b 0.05) and vascular disease (22.0% vs. 12.4%, p b 0.05).

Therapeutic intervention

There was no difference between the 2 groups in term of intrave- nous thrombolysis. However, patients in group 1 were more frequently treated by mechanical thrombectomy (28.3% vs. 11.2%, p b 0.05).

Paraclinical results

Biomarkers levels (BNP, troponin I and D-dimers) were significantly higher in group 1, with a frequency of increase being the most impor- tant for BNP (BNP 83.0% vs. 39.2%, D-dimers 65.4% vs. 44.22%, troponin I 24.5% vs. 8.6%, p b 0.05 for the 3 parameters). Frequency of multifocal stroke did not differ between the groups.

Parameters associated with a cardioembolic mechanism (C1)

Results from univariate and multivariate analysis are presented in Table 2. Among clinical data recorded at admission only initial stroke

Fig. 1. Patient flow chart.

severity (OR 1.04, 95 CI% 1.004-1.07, p b 0.05) and a history of heart fail- ure (OR 3.03, 95% CI 1.19-7.73, p b 0.05) were associated with a pure C1 mechanism at 3 months. ECG abnormalities were also significantly asso- ciated with a C1 mechanism in both models while multiple stroke loca- tion was not. Among the 3 biological markers, BNP was the only one that reached statistical significance in the multivariate analysis in model 1 (OR 4.34, 95% CI 2.59-7,29, p b 0.05). When the sum of the 3 bio- markers was studied in model 2, it was associated with a cardioem- bolic mechanism (OR 2.37, 95% CI 1.86-3.02, p b 0.05). Sensitivity and specificity for each biological marker elevation were respec- tively: 83.0% and 60.8% for BNP, 24.5% and 91.4% for troponin, and

65.4% and 55.7% for D-dimers.

Two additional multivariate analyses were carried out (data not shown): the first studied markers associated to patients ranked C1 only due to AF (model 3), the second studied markers associated with AF while excluding patients whose AF was already known before the is- chemic stroke (model 4). In model 3, BNP level and ECG abnormalities were the only parameters associated to AF, with BNP remaining the most significantly (respectively OR 5.65, 95% CI 3.25-9.81, p b 0.05 and OR 1.59, 95% CI 1.17-2.17, p b 0.05). In model 4, BNP was the only parameter associated with newly identified AF (OR 4.70, 95% CI 2.12- 10.42, p b 0.05).


The main results of this study are the following: i) Increased levels of BNP, troponin and D-dimers early after stroke are significantly associ- ated with cardiac sources of stroke, mainly AF; ii) Increased BNP is the most sensitive biomarker compared to troponin and D-dimers and iii) Biological markers are more associated with cardiac stroke mechanism than clinical and radiological markers represented by the CHADS-VASC score and multifocal stroke distribution.

Several studies have investigated the association between biological markers such as BNP, troponin or D-dimers and the underlying stroke mechanism [10,14-17]. However few of them have assessed the predic- tive value of a combination of these measures that encompass different pathophysiological processes such as cardiac insufficiency, ischemic heart disease and hypercoagulability. Our results confirm the strong as- sociation between a high global biological score of the measured bio- markers and the probability of a cardiac source of stroke (OR 2.37. 95% CI 1.86-3.02) but do not demonstrate a better association than BNP alone (OR 4.34. 95% CI 2.59-7.29. p b 0.05) which do not support the hy- pothesis that the measure of these 3 markers might improve the early identification of patients with a probable cardiac stroke mechanism in routine clinical practice. BNP is mainly synthesized by cardiomyocytes

Table 1

Characteristics of stroke patients according to stroke mechanism.

practice could offer an interesting opportunity to identify patients with a high probability of cardiac stroke mechanism. Although our results

Total population

Group 1 (C1


Group 2 (non-C1 patients)

Comparison between groups 1 and 2

confirmed the well-known association between cardio-embolic mecha- nism and more severe stroke [7], we failed to identify a significant asso- ciation with the CHADS-VASC status while it has been reported [23].

n = 634 n = 159 n = 475 p

Demographical and clinical data

Female (%)

249 (39.3)

70 (44.0)

179 (37.7)


Age (SD)



67.1 (16.9)




Pre-stroke mRS

0 (0-5)

0 (0-5)

0 (0-4)


Admission NIHSS (SD)

5.4 (6.2)

7.8 (7.0)

4.5 (5.7)


mRS at discharge

1 (0-6)

1 (0-6)

1 (0-6)


Cardiovascular comorbidities

Diabetes (%)

161 (25.4)

42 (26.4)

119 (25.1)


Hypertension (%)

376 (59.3)


268 (56.4)



Cardiac insufficiency (%)

26 (4.1)

16 (10.1)

10 (2.1)


Cerebral ischemia (%)

87 (13.7)

22 (13.8)

65 (13.7)


Vascular disease (%)

94 (15.2)

35 (22.0)

59 (12.4)


Atrial fibrillation (%)

90 (14.2)

70 (44.0)

20 (4.2)



3 (0-7)

3 (0-7)

3 (0-7)



IV thrombolysis (%)

125 (19.7)

35 (22.0)

90 (18.9)



98 (15.5)

45 (28.3)

53 (11.2)


thrombectomy (%) Imaging results

Multifocal stroke (%) 113 (17.8) 30 (18.9) 83 (17.5) 0.69

ECG results

Normal 500 (80.4) 85 (53.5) 415 (87.4) b0.05

Atrial fibrillation

71 (11.4)

56 (35.2)

15 (3.2)

severe stroke, high BNP level and Abnormal ECG can help to identify a

Other abnormalities

51 (8.2)

14 (8.8)

37 (7.79)

subpopulation of stroke patients with a high probability of cardiac

This result might be explained by the predominant role played by heart failure which is highly correlated to increased BNP levels [24]. However, we also found an independent association between higher BNP at admission and the identification of AF during follow-up suggest- ing that this marker is not the sole consequence of cardiac insufficiency but could be considered as an indicator of a myocardial dysfunction in- creasing the risk of AF [25]. This hypothesis is also supported by the as- sociation observed between any ECG abnormalities and C1 stroke mechanism. The combination of clinical, biological and abnormal car- diac rhythm abnormalities has already been reported to improve the detection of patients with a high probability of cardiac stroke mecha- nism [26] and is reinforced by the present results. Conversely, stroke lo- cation was a poor indicator, in line with previous results [27].

Results of this study have to be interpreted cautiously due to several limitations. First a high number of patients had neither TEE nor long term ECG monitoring which might have led to overestimate the propor- tion of patients with cryptogenic stroke. Second, echocardiographic pa- rameters were not included in the model while several studies have demonstrated a clear association between morphological abnormalities such as left atrial dilation and a higher risk of AF, therefore improving predictive models when added to clinical, biological and heart rhythm abnormalities [6]. However access to TEE is still limited at many centers in the early stroke phase and our results suggest that a combination of

Biological results

Increased troponin I (%) 80 (12.6) 39 (24.5) 41 (8.6) b0.05

stroke sources. Such parameters might be used to improve the selection of patients for whom further cardiologic investigations such as continu-

Increased D-dimers (%) 314 (49.5) 104


Increased BNP (%) 318 (50.2) 132


Statistically significant p value (b0.05) appears in bold.

210 (44.2) b0.05

186 (39.2) b0.05

ous long term ECG monitoring would be the most useful. The recent in- terruption of the Navigate Trial for futility reinforces the need to improve our ability to identify patients with cardiac stroke sources in the context of limited access to advanced cardiac investigation in our daily practice.

from the ventricles and is increased in most cases of cardiac dysfunction contributing to AF. Our results are in accordance with several previous studies and meta-analysis that have established an association between increased BNP or NT-proBNP and cardioembolic stroke [18-21]. How- ever, results were better for NT-proBNP [22] which is a marker not available in every center for clinical practice at a low cost. Herein, our re- sults suggest that a measurement of BNP as it is easily obtained in daily


  1. Arauz A, Morelos E, Colin J, Roldan J, Barboza MA. Comparison of functional outcome and stroke recurrence in patients with embolic stroke of undetermined source (ESUS) vs. cardioembolic stroke patients. PLoS One 2016;11(11):e0166091.
  2. van Walraven C, Hart RG, Singer DE, Laupacis A, Connolly S, Petersen P, et al. Oral an- ticoagulants vs aspirin in nonvalvular atrial fibrillation: an individual patient meta- analysis. JAMA 2002 Nov 20;288(19):2441-8.

    Table 2

    Acute stroke parameters associated with diagnosis of cardioembolic mechanism at 3 months (C1).

    Clinical and paraclinical data Univariate analysis Multivariate analysis model 1 Multivariate analysis model 2


    95% CI



    95% CI



    95% CI















    Admission NIHSS























    Heart failure







    Cerebral ischemia







    Vascular disease













    Multifocal stroke


















    Elevated troponin I







    Elevated D-dimers







    Elevated BNP







    Global biological score



    Statistically significant p value (b0.05) appears in bold.

    Kolominsky-Rabas PL, Weber M, Gefeller O, Neundoerfer B, Heuschmann PU. Epide- miology of ischemic stroke subtypes according to TOAST criteria: incidence, recur- rence, and long-term survival in ischemic stroke subtypes: a population-based study. Stroke 2001 Dec 1;32(12):2735-40.

  3. Ntaios G, Papavasileiou V, Lip GYH, Milionis H, Makaritsis K, Vemmou A, et al. Em- bolic stroke of undetermined source and detection of atrial fibrillation on follow- up: how much causality is there? J Stroke Cerebrovasc Dis Off J Natl Stroke Assoc 2016 Dec;25(12):2975-80.
  4. Wachter R, Groschel K, Gelbrich G, Hamann GF, Kermer P, Liman J, et al. Holter-elec- trocardiogram-monitoring in patients with acute ischaemic stroke (Find- AFRANDOMISED): an open-label randomised controlled trial. Lancet Neurol 2017 Apr;16(4):282-90.
  5. Levy S, Maarek M, Coumel P, Guize L, Lekieffre J, Medvedowsky JL, et al. Characteri- zation of different subsets of atrial fibrillation in General practice in France: the ALFA study. The College of French Cardiologists. Circulation 1999 Jun 15;99(23):3028-35.
  6. Llombart V, Antolin-Fontes A, Bustamante A, Giralt D, Rost NS, Furie K, et al. B-type Natriuretic peptides help in cardioembolic stroke diagnosis: pooled data meta-anal- ysis. Stroke 2015 May;46(5):1187-95.
  7. Scheitz JF, Erdur H, Haeusler KG, Audebert HJ, Roser M, Laufs U, et al. Insular cortex lesions, cardiac troponin, and detection of previously unknown atrial fibrillation in acute ischemic stroke: insights from the Troponin elevation in acute ischemic stroke study. Stroke 2015;46(5):1196-201.
  8. Ward F, McGovern R, Cotter PE. Troponin-I is a predictor of a delayed diagnosis of atrial fibrillation in acute ischemic stroke and transient ischemic attack. J Stroke Cerebrovasc Dis 2015 Jan;24(1):66-72.
  9. Liu L-B, Li M, Zhuo W-Y, Zhang Y-S, Xu A-D. The role of Hs-CRP, D-dimer and fibrin- ogen in differentiating etiological subtypes of ischemic stroke. PLoS One 2015 Feb 13;10(2):e0118301 (Ai J, editor).
  10. Camm AJ, Lip GYH, De Caterina R, Savelieva I, Atar D, Hohnloser SH, et al. 2012 fo- cused update of the ESC Guidelines for the management of atrial fibrillation: an up- date of the 2010 ESC Guidelines for the management of atrial fibrillation. Developed with the special contribution of the European Heart Rhythm Association. Eur Heart J 2012 Nov;33(21):2719-47.
  11. Righini M, Van Es J, Den Exter PL, Roy P-M, Verschuren F, Ghuysen A, et al. Age-ad- justed D-dimer cutoff levels to rule out pulmonary embolism: the ADJUST-PE study. JAMA 2014 Mar 19;311(11):1117.
  12. Amarenco P, Bogousslavsky J, Caplan LR, Donnan GA, Wolf ME, Hennerici MG. The ASCOD phenotyping of ischemic stroke (Updated ASCO Phenotyping). Cerebrovasc Dis Basel Switz 2013;36(1):1-5.
  13. Zi W-J, Shuai J. Plasma D-dimer levels are associated with stroke subtypes and in- farction volume in patients with acute ischemic stroke. PLoS One 2014 Jan 20;9

    (1):e86465 (Ai J, editor).

    Chaudhuri JR, Sharma VK, Mridula KR, Balaraju B, Bandaru VCSS. Association of plasma brain natriuretic peptide levels in acute ischemic stroke subtypes and out- come. J Stroke Cerebrovasc Dis Off J Natl Stroke Assoc 2015 Feb;24(2):485-91.

  14. Isenegger J, Meier N, Lammle B, Alberio L, Fischer U, Nedeltchev K, et al. D-dimers predict stroke subtype when assessed early. Cerebrovasc Dis Basel Switz 2010;29 (1):82-6.
  15. Alvarez-Perez FJ, Castelo-Branco M, Alvarez-Sabin J. Usefulness of measurement of fibrinogen, D-dimer, D-dimer/fibrinogen ratio, C reactive protein and erythrocyte sedimentation rate to assess the pathophysiology and mechanism of ischaemic stroke. J Neurol Neurosurg Psychiatry 2011 Sep;82(9):986-92.
  16. Sanak D, Hutyra M, Kral M, Bartkova A, Zapletalova J, Fedorco M, et al. Paroxysmal atrial fibrillation in young cryptogenic ischemic stroke: a 3-week ECG Holter moni- toring study. Biomed Pap Med Fac Univ Palacky Olomouc Czechoslov 2015 Jun;159 (2):283-7.
  17. Sonderer J, Katan Kahles M. Aetiological blood biomarkers of ischaemic stroke. Swiss Med Wkly 2015;145:w14138.
  18. Shibazaki K, Kimura K, Aoki J, Sakai K, Saji N, Uemura J. Plasma brain natriuretic pep- tide as a predictive marker of early recurrent stroke in cardioembolic stroke patients. J Stroke Cerebrovasc Dis Off J Natl Stroke Assoc 2014 Dec;23(10):2635-40.
  19. Wu Z, Zhao M, He M, Zeng H, Tan F, Li K, et al. Validation of the use of B-type natri- uretic peptide Point-of-care test platform in preliminary recognition of cardioem- bolic stroke patients in the ED. Am J Emerg Med 2015 Apr;33(4):521-6.
  20. Yang X, Gao S, Ding J, Chen Y, Zhou X, Wang J-E. Plasma D-dimer predicts short-term poor outcome after acute ischemic stroke. PLoS One 2014 Feb 24;9(2):e89756 (Salluh JIF, editor).
  21. Friberg L, Rosenqvist M, Lindgren A, Terent A, Norrving B, Asplund K. High preva- lence of atrial fibrillation among patients with ischemic stroke. Stroke 2014 Sep;45 (9):2599-605.
  22. Maisel AS, Krishnaswamy P, Nowak RM, McCord J, Hollander JE, Duc P, et al. Rapid measurement of B-type natriuretic peptide in the emergency diagnosis of heart fail- ure. N Engl J Med 2002 Jul 18;347(3):161-7.
  23. Knudsen CW, Omland T, Clopton P, Westheim A, Wu AHB, Duc P, et al. Impact of atrial fibrillation on the diagnostic performance of B-type natriuretic peptide con- centration in dyspneic patients: an analysis from the breathing not properly multi- national study. J Am Coll Cardiol 2005 Sep 6;46(5):838-44.
  24. Suissa L, Bresch S, Lachaud S, Mahagne MH. Brain natriuretic peptide: a relevant marker to rule out delayed atrial fibrillation in stroke patient. J Stroke Cerebrovasc Dis Off J Natl Stroke Assoc 2013 Oct;22(7):e103-10.
  25. Bernstein RA, Di Lazzaro V, Rymer MM, Passman RS, Brachmann J, Morillo CA, et al. Infarct topography and detection of atrial fibrillation in cryptogenic stroke: results from CRYSTAL AF. Cerebrovasc Dis Basel Switz 2015;40(1-2):91-6.

Leave a Reply

Your email address will not be published. Required fields are marked *