ARTICLE CLASS OF EVIDENCE

Development and validation of a score to detect

paroxysmal atrial fibrillation after stroke
Timo Uphaus, MD, Mark Weber-Krüger, MD, Martin Grond, MD, Gerrit Toenges, MSc, Correspondence
Antje Jahn-Eimermacher, PhD, Marek Jauss, MD, Paulus Kirchhof, MD, Rolf Wachter, MD,* Dr. Gröschel
and Klaus Gröschel, MD* klaus.groeschel@

®
unimedizin-mainz.de
Neurology 2019;92:e115-e124. doi:10.1212/WNL.0000000000006727

Abstract RELATED ARTICLE

Objective Editorial
Prolonged monitoring times (72 hours) are recommended to detect paroxysmal atrial fibril- ECG monitoring after
lation (pAF) after ischemic stroke but this is not yet clinical practice; therefore, an individual acute ischemic stroke: Does
patient selection for prolonged ECG monitoring might increase the diagnostic yield of pAF in patient selection matter?
a resource-saving manner. Page 65

Methods MORE ONLINE

We used individual patient data from 3 prospective studies (ntotal = 1,556) performing pro-
Class of Evidence
longed Holter-ECG monitoring (at least 72 hours) and centralized data evaluation after TIA or
Criteria for rating
stroke in patients with sinus rhythm. Based on the TRIPOD (Transparent Reporting of therapeutic and diagnostic
a Multivariable Prediction Model for Individual Prognosis or Diagnosis) guideline, a clinical studies
score was developed on one cohort, internally validated by bootstrapping, and externally
NPub.org/coe
validated on 2 other studies.

Results Podcast
pAF was detected in 77 of 1,556 patients (4.9%) during 72 hours of Holter monitoring. After Dr. Andy Southerland talks
logistic regression analysis with variable selection, age and the qualifying stroke event (cate- with Dr. Timo Uphaus
gorized as stroke severity with NIH Stroke Scale [NIHSS] score ≤5 [odds ratio 2.4 vs TIA; 95% about his paper on the
development and
confidence interval 0.8–6.9, p = 0.112] or stroke with NIHSS score >5 [odds ratio 7.2 vs TIA;
validation of a score to
95% confidence interval 2.4–21.8, p < 0.001]) were found to be predictive for the detection of
detect paroxysmal atrial
pAF within 72 hours of Holter monitoring and included in the final score (Age: 0.76 points/
fibrillation after stroke.
year, Stroke Severity NIHSS ≤5 = 9 points, NIHSS >5 = 21 points; to Find AF [AS5F]). The
NPub.org/4t0r7p
high-risk group defined by AS5F is characterized by a predicted risk between 5.2% and 40.8%
for detection of pAF with a number needed to screen of 3 for the highest observed AS5F points
within the study population. Regarding the low number of outcomes before generalization of
AS5F, the results need replication.

Conclusion
The AS5F score can select patients for prolonged ECG monitoring after ischemic stroke to
detect pAF.

Classification of evidence
This study provides Class I evidence that the AS5F score accurately identifies patients with
ischemic stroke at a higher risk of pAF.

*These authors contributed equally to this work.

From the Department of Neurology (T.U., K.G.), and Institute of Medical Biostatistics, Epidemiology and Informatics (G.T., A.J.-E.), University Medical Center of the Johannes
Gutenberg University Mainz; Department of Cardiology and Pneumology (M.W.-K.), University of Göttingen; Clinic and Policlinic for Cardiology (R.W.), University Hospital Leipzig,
Germany; Department of Neurology (M.G.), Kreisklinikum Siegen; Darmstadt University of Applied Sciences (A.J.-E.); Department of Neurology (M.J.), Hainich Klinikum, Mühlhausen,
Germany; Institute of Cardiovascular Sciences (P.K.), University of Birmingham; and Department of Cardiology (P.K.), SWBH and UHB NHS Trusts, Birmingham, UK.

Go to Neurology.org/N for full disclosures. Funding information and disclosures deemed relevant by the authors, if any, are provided at the end of the article.

Copyright © 2018 American Academy of Neurology e115

Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
 Glossary
AF = atrial fibrillation; AS5F = Age, Stroke Severity NIHSS >5 to Find AF; AUC = area under the curve; CHADS2 = congestive
heart failure, hypertension, age >75, diabetes mellitus, stroke/TIA; CRYSTAL-AF = Cryptogenic Stroke and Underlying AF;
ESUS = Embolic Stroke of Undetermined Source; IDEAS = Ideas to Detect Atrial Fibrillation in Stroke Patients Study; IQR =
interquartile range; NIHSS = NIH Stroke Scale; NNS = number needed to screen; NRI = net reclassification improvement;
OR = odds ratio; pAF = paroxysmal atrial fibrillation; ROC = receiver operating characteristic curve; TRIPOD = Transparent
Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis.

Atrial fibrillation (AF) remains one of the most important
causes of acute ischemic stroke and is associated with an
increased mortality, severe disability, and high stroke re-
currence.1 Early detection of silent, undiagnosed AF is
therefore a public health priority.2–4 Detection of AF in
patients with stroke has immediate effects on treatment as oral
anticoagulation is recommended in all patients with AF and
a history of ischemic stroke.5,6 While chronic forms of AF are
easily diagnosed by 12-lead ECG, paroxysmal AF (pAF) fre-
quently escapes routine diagnostic workup because of its in-
termittent behavior, appearance within clusters, and its
clinically asymptomatic nature. Detection of pAF increases
when ECG monitoring times are prolonged in stroke survi-
vors,7 leading to a recommendation for ECG monitoring of at
least 72 hours in these patients.3 Yet, less than 20% of patients
presenting with ischemic stroke underwent ECG monitoring
for more than 24 hours in recent surveys,8,9 most likely be-
cause of limited availability of the technical and human
resources needed for prolonged ECG monitoring and evalu-
ation. Hence, identification of stroke survivors at particularly
high risk of pAF might support a wider and efficient use of
prolonged ECG monitoring in patients with ischemic stroke.
The aim of the current study was to develop and validate a risk
score for detection of pAF in patients after ischemic stroke with
subsequent 72 hours of minimum Holter-ECG monitoring.
IDEAS, the Find-AF, and the Find-AFrandomized cohort. Patients
with lacunar infarcts and lacunar syndromes were not excluded.
Patients with a severe ipsilateral carotid or intracranial artery
stenosis or at an age younger than 60 years were excluded within
the Find-AFrandomized trial. IDEAS10 was a prospective, multi-
center, observational study including 1,135 patients after ische-
mic stroke or TIA who all underwent 72-hour Holter-ECG in
addition to routine pAF detection practice. Find-AF11 was
a prospective monocentric study that enrolled 224 patients after
ischemic stroke or TIA with sinus rhythm on admission ECG;
7-day Holter-ECG was performed in all of these patients. Three
patients were excluded because of missing NIH Stroke Scale
(NIHSS) scores, leaving 221 patients for analysis. Find-
AFrandomized7 was a prospective, multicenter trial randomizing
patients with ischemic stroke and symptom onset within 7 days
before hospital admission and sinus rhythm on admission ECG
into either prolonged monitoring or standard-of-care monitor-
ing. Two hundred patients were randomized to the intervention
group with prolonged monitoring and received 10 days of
Holter-ECG directly after the hospital admission and again after
3 and 6 months.
The ECG data from all Holter monitorings were analyzed in
a central ECG core laboratory under the supervision of an
experienced electrophysiologist (IDEAS: P.K.; Find-AF
studies: R.W.).

Standard protocol approvals, registrations,

Methods
Primary research question
Development and validation of a clinical risk score for detection
of pAF in patients after ischemic stroke with subsequent
72-hour minimum Holter-ECG monitoring: classification of
evidence I.
Study population
This analysis included the individual patient data of 3 prospective
studies investigating the detection of pAF with prolonged Holter
monitoring times, namely, the Ideas to Detect Atrial Fibrilla-
tion in Stroke Patients Study (IDEAS),10 Find-AF (ISRCTN
46104198),11 and Find-AFrandomized (NCT01855035),7 which
all focused on survivors of acute ischemic stroke or TIA. Con-
cerning the inclusion and exclusion criteria, the AS5F score was
developed in an almost unselected patient cohort irrespective of
the suspected cause of stroke. In addition, a history of AF and
documented AF on admission were exclusion criteria for the
and patient consents
All 3 studies included received an approval by an ethical stand-
ards committee on human experimentation. Written informed
consent was obtained from all patients participating in the study.
Definitions of variables
AF was defined as at least one period of >30 seconds of
an absolute arrhythmia without detectable P waves.3,7 For
calculation of the CHADS2 (congestive heart failure, hyper-
tension, age >75, diabetes mellitus, stroke/TIA) score,12 in-
formation on each patient’s medical history was taken from
the individualized study case report forms. Arterial hyper-
tension was defined as treatment with antihypertensive drugs
or either systolic blood pressure above 140 mm Hg or di-
astolic blood pressure above 90 mm Hg.
Primary outcomes
The 3 studies investigated different intervals of prolonged
monitoring duration (IDEAS: 3 days; Find-AF: 7 days; and

e116 Neurology | Volume 92, Number 2 | January 8, 2019 Neurology.org/N

Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
Find-AFrandomized: 10 days performed 3 times) to detect AF. For
consistency, we only included AF detected within the first 3 days
of monitoring for development and validation of the risk score.
Statistical analyses
For development, validation, and reporting of the proposed
risk score, we followed the TRIPOD (Transparent Reporting
of a Multivariable Prediction Model for Individual Prognosis
or Diagnosis) guideline13 (see TRIPOD checklist; data
available from Dryad, table 1, doi.org/10.5061/dryad.
ms3m3n2). Patient data were split into a dataset for score
development (IDEAS, n = 1,135) and a dataset for external
validation (Find-AF and Find-AFrandomized, n = 421). Sex,
coronary heart disease, age, and the severity of the preceding
ischemic event were considered as potential predictors and
selected on clinical basis and after univariate analysis of the
IDEAS dataset. The score was developed in the IDEAS cohort
using a logistic regression analysis with a backward variable
selection algorithm that was performed using the mfp package
(version 1.5.2) in R. To facilitate the application of the model,
a scoring system was developed based on the regression
coefficients of the selected model.13
The apparent performance of the final model was assessed on
the IDEAS data in terms of discrimination (receiver operating
characteristic curve–area under the curve [ROC-AUC]) and
calibration (intercept and slope of the calibration line). To
adjust for optimism due to overfitting, we internally validated
these performance measures on 1,000 bootstrap samples of
the IDEAS data. In addition, we externally validated the
model’s performance in the combined Find-AF and Find-
AFrandomized datasets. The performance of the score was ex-
amined for the 72-hour outcome as well as for the whole study
outcome in Find-AF (7 days) and Find-AFrandomized (10 days)
as a sensitivity analysis.
The score was used to build a classifier discriminating be-
tween patients with low and high risk of experiencing pAF by
applying a cutoff value that maximized the Youden index in
the IDEAS cohort. For comparison, a classifier based on the
CHADS2 score and one based on a logistic regression model
with age as the only covariate were derived in the same
manner. The performance of these 3 classifiers was compared
by means of the net reclassification improvement (NRI) on
the validation cohort (FIND-AF and Find-AFrandomized). The
p values for NRIs were calculated based on an asymptotic test
proposed by Pencina et al.13
We applied the Mann-Whitney U test, Student t test, χ 2 test,
or Fisher exact test as appropriate to examine differences in
the mean and proportions between 2 samples. To compare
the areas under 2 correlated ROC curves, the nonparametric
test by DeLong et al.14 was used with a 2-sided alternative. No
adjustments for multiple testing were performed. The p values
are given for descriptive purposes only. Because of the large
number of tests, p values should be interpreted with caution
and in connection with effect estimates.
Neurology.org/N
Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
All statistical analyses were performed with the software R
(version 3.3.1). For details, see data available from Dryad
(Methods, doi.org/10.5061/dryad.ms3m3n2).
Data availability
Data that support the findings of this study are available from
the corresponding author on request from any qualified
investigator.
Results
Patient characteristics
Data from a total of 1,556 individual patients from 3 in-
dependent prospective studies examining the diagnostic yield
of prolonged Holter-ECG monitoring were ultimately ana-
lyzed (median age 70 years, interquartile range [IQR] 60–76,
55.5% male; for details see table 1). pAF was found in 77
patients (4.9%) within the first 72 hours, whereas beyond 72
hours (Find-AF 7 days, Find-AFrandomized 10 days), pAF could
be detected in an additional 19 patients (96 patients to-
tal, 6.2%).
Most patients presented with an ischemic stroke (1,214, 78%)
in contrast to only transient symptoms (TIA; 342, 22%).
Distributions of cardiovascular risk factors, baseline NIHSS
scores, the CHADS2 score, and further patient characteristics
are depicted in table 1.
Patients diagnosed with pAF within 72 hours by prolonged
Holter-ECG monitoring (table 2) had a higher median age
(pAF detection 78 [IQR 72–81] vs no pAF detection 70 [IQR
60–76], p < 0.001) and presented more frequently with
symptoms of an ischemic stroke in contrast to transient
symptoms (pAF detection 92.2% vs no pAF detection 77.3%,
p = 0.002). In addition, stroke severity measured by the
NIHSS was significantly higher in patients with pAF (pAF
detection 4 [IQR 2–9] vs no pAF detection 2 [IQR 1–4], p <
0.001). Patients with pAF more often experienced coronary
artery disease (pAF detection 26% vs no pAF detection
15.4%, p = 0.014), arterial hypertension (pAF detection
77.9% vs no pAF detection 65.3%, p = 0.023), had a higher
CHADS2 score (pAF detection 4 vs no pAF detection 3, p <
0.001), and were less frequently current or former smokers
(pAF detection 32.5% vs no pAF detection 45.7%, p = 0.017).
Details regarding patients diagnosed with pAF in the IDEAS
cohort can be found in the original publication by Grond
et al.10
Score development and validation
For development and validation of the proposed risk score, we
considered the TRIPOD guideline13 (data available from
Dryad, table 1, doi.org/10.5061/dryad.ms3m3n2), resulting
in the logistic regression model shown in table 3. For score
development, we considered all variables with p < 0.1 in the
univariable analysis of the IDEAS cohort and known risk
factors for pAF (sex, coronary heart disease, age, and severity
Neurology | Volume 92, Number 2 | January 8, 2019 e117
 Table 1 Baseline characteristics of the study population
All IDEAS Find-AF Find-AFrandomized

No. 1,556 1,135 221 200

Age, y 70 (60–76) 69 (59–76) 70 (60–79) 72 (65–77)

Male 863 (55.5) 621 (54.7) 128 (57.9) 114 (57.0)

BMI, kg/m2 27.5 ± 4.9 27.4 ± 4.9 27.5 ± 5.7 27.5 ± 4.3

AF detected, within 72 h 77 (4.9) 49 (4.3) 20 (9.0) 8 (4.0)

AF detected, within study period 96 (6.2) 49 (4.3) 29 (13.1) 18 (9.0)

Stroke 1,214 (78) 866 (76.3) 148 (67.0) 200 (100)

TIA 342 (22) 269 (23.7) 72 (33.0) 0 (0)

Cardiovascular risk factors

Diabetes mellitus 336 (21.6) 231 (20.4) 49 (22.2) 56 (28.0)

Hyper-/dyslipidemia 556 (35.7) 402 (35.4) 77 (34.8) 77 (38.5)

Arterial hypertension 1,025 (65.9) 706 (62.2) 162 (73.3) 157 (78.5)

Smoking (current/former) 702 (45.1) 558 (49.2) 54 (24.4) 90 (45.0)

Previous stroke 280 (18.0) 198 (17.4) 34 (15.4) 48 (24.0)

CHF 88 (5.7) 66 (5.8) 11 (5.0) 11 (5.5)

Coronary heart disease 248 (15.9) 188 (16.6) 33 (14.9) 27 (13.5)

Peripheral arterial disease 66 (4.2) 46 (4.1) 6 (2.7) 14 (7.0)

NIHSS score, baseline 2 (1–4) 2 (0–4) 3 (1–5) 3 (1–5)

CHADS2 score 3 (3–4) 3 (3–4) 3 (2–4) 3 (3–4)

Abbreviations: AF = atrial fibrillation; BMI = body mass index; CHADS2 = congestive heart failure, hypertension, age >75, diabetes mellitus, stroke/TIA; CHF =
congestive heart failure; IDEAS = Ideas to Detect Atrial Fibrillation in Stroke Patients Study; NIHSS = NIH Stroke Scale.
Values are expressed as n (%), mean ± SD, or median (25th–75th percentile).

of the qualifying stroke event; TIA as reference, stroke with
NIHSS score > 5, stroke with NIHSS score ≤ 5). Defining the
severity using an NIHSS cutoff of 5 was an initial step that
maximized the ROC-AUC in univariate logistic regression on
the IDEAS data. The backward variable selection algorithm
on the IDEAS data resulted in a logistic regression model that
included age in years (odds ratio [OR] 1.07, p < 0.001) and
the qualifying stroke event (stroke with NIHSS ≤5 vs TIA:
OR 2.378, p = 0.112; stroke with NIHSS >5 vs TIA: OR 7.23,
p < 0.001) but not sex and coronary heart disease for pre-
diction of pAF detection within the first 72 hours. Details are
provided in table 3. The final model for calculation of in-
dividual risk prediction is given in data available from Dryad
(figure 1, doi.org/10.5061/dryad.ms3m3n2). These results
led to the final risk score name of AS5F (Age, Stroke Severity
NIHSS >5 to Find AF; table 4).
The score’s apparent performance on the IDEAS data (ROC-
AUC 0.78, calibration line intercept <0.01, and slope 0.97)
only slightly differed from the adjusted measures obtained in
the internal validation step (ROC-AUC 0.78, calibration line
intercept <0.01, and slope 0.86), reflecting the internal val-
idity of AS5F. Moreover, the score was shown to keep its
discriminative performance (ROC-AUC 0.75) and predictive
accuracy (calibration line intercept <0.01 and slope 1.06) in
an external validation using the 72-hour outcome of the Find-
AF and Find-AFrandomized studies. See data available from
Dryad (figure 2, doi.org/10.5061/dryad.ms3m3n2).
The total points of the AS5F score can be converted into
a predicted risk for pAF detection by the formula given in data
available from Dryad (figure 1, doi.org/10.5061/dryad.
ms3m3n2); this is further illustrated in figure 1A. To easily
convert AS5F score points into the predicted pAF risk, a no-
mogram, displayed in figure 1B, was created. For example,
a score of 83 total points for an 82-year-old patient with stroke
who has an NIHSS score >5 reflects an approximate 19%
predicted risk of pAF detection within 72 hours of Holter-
ECG monitoring, which is reflected by a number needed to
screen (NNS) of 5 patients to detect 1 patient with pAF
within a 72-hour Holter-ECG (figure 1B, red line).
AS5F discriminates between low- and high-
risk patients
To select patients for prolonged monitoring after ischemic
stroke, we classified them as either at low or high risk of pAF

e118 Neurology | Volume 92, Number 2 | January 8, 2019 Neurology.org/N

Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
Table 2 Characteristics of patients detected with pAF during 72 hours of Holter monitoring
pAF detection (72 h Holter-ECG) No pAF detection (72 h Holter-ECG) p Value

No. 77 1,479

Age, y 78 (72–81) 70 (60–76) <0.001

Male 38 (49.4) 825 (55.8) 0.268

BMI, kg/m2 27 ± 3 27 ± 5 0.345

Stroke (vs TIA) 71 (92.2) 1,143 (77.3) 0.002

Cardiovascular risk factors

Diabetes mellitus 21 (27.3) 315 (21.3) 0.222

Hyper-/dyslipidemia 30 (39.0) 526 (35.6) 0.496

Arterial hypertension 60 (77.9) 965 (65.3) 0.023

Smoking, current/former 25 (32.5) 677 (45.7) 0.017

Previous stroke 13 (16.9) 267 (18.1) 0.792

Congestive heart failure 6 (7.8) 82 (5.5) 0.405

Coronary heart disease 20 (26.0) 228 (15.4) 0.014

Peripheral arterial disease 4 (5.2) 62 (4.2) 0.670

NIHSS score, baseline 4 (2–9) 2 (1–4) <0.001

TOAST classification <0.001

Large artery atherosclerosis 17 (22.2) 447 (30.2)

Cardioembolism 36 (46.8) 213 (14.4)

Small vessel occlusion 9 (11.7) 235 (15.9)

Stroke of other identified cause 1 (1.3) 68 (4.6)

Stroke of unknown cause 14 (18.2) 516 (34.9)

CHADS2 score 4 (3–4) 3 (3–4) <0.001

Abbreviations: BMI = body mass index; CHADS2 = congestive heart failure, hypertension, age >75, diabetes mellitus, stroke/TIA; NIHSS = NIH Stroke Scale;
pAF = paroxysmal atrial fibrillation; TOAST = Trial of Org 10172 in Acute Stroke Treatment.
Values are expressed as n (%), mean ± SD, or median (25th–75th percentile).

detection by using the AS5F score that maximized the to an NNS of fewer than 20 patients to detect 1 patient with
Youden index on the IDEAS cohort as a cutoff. The AS5F pAF during 72 hours of Holter-ECG monitoring at the
threshold was found to be 67.5 points, which reflects lower limit in the high-risk group. For more information,
a predicted risk of pAF detection of 5.2%. This corresponds see data available from Dryad, figure 3, doi.org/10.5061/
dryad.ms3m3n2; refer to unimedizin-mainz.de/neuro-
logie/header/as5f.html for an easy-to-use online calculator
of the score. Regarding the discrimination between low- and
Table 3 Logistic regression analysis results for high-risk patients, the AS5F score performed significantly
paroxysmal atrial fibrillation detection within 72
better on the 72-hour outcome of the validation cohort
hours derived from the IDEAS data
compared to the CHADS2 score (NRI 0.22 with p = 0.047;
Variable Odds ratio 95% CI p Value ROC-AUC 0.75 for AS5F and 0.61 for CHADS2, p = 0.0032;
Age 1.07 1.04–1.11 <0.001 figure 2A). Comparing the performance of AS5F to age alone
as continuous variable, we observed a larger AUC for AS5F
Stroke with NIHSS ≤5 vs TIA 2.38 0.81–6.94 0.112
within a period of 72 hours of ECG monitoring (NRI 0.09 with
Stroke with NIHSS >5 vs TIA 7.23 2.40–21.76 <0.001 p = 0.36; ROC-AUC 0.75 for AS5F and 0.69 for age as con-
tinuous variable, p = 0.09). This difference becomes even more
Abbreviations: CI = confidence interval; IDEAS = Ideas to Detect Atrial Fi-
brillation in Stroke Patients Study; NIHSS = NIH Stroke Scale.
relevant when considering AS5F vs age alone for ECG moni-
toring intervals beyond 72 hours (NRI 0.22 with p = 0.004;

Neurology.org/N Neurology | Volume 92, Number 2 | January 8, 2019 e119

Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.

Table 4 Acronym AS5F (Age and Stroke Severity NIHSS >5
to Find AF) and calculation of the risk score
Age × 0.76 —
Stroke severity NIHSS ≤5 9 points
Stroke severity NIHSS >5 21 points
AS5F score Total points
Examples of AS5F predicted risk of pAF detection within 72 hours of
Holter-ECG
Stroke
severity, Stroke severity, + pAF
Age, y TIA NIHSS ≤5 stroke NIHSS >5 AS5F risk NNS
46 X 35 0.25 400
71 X 75 10 10
82 X 83 19 5 use of anticoagulative medication in most cases, which is one
Abbreviations: NIHSS = NIH Stroke Scale; NNS = number needed to screen;
pAF = paroxysmal atrial fibrillation.
See web page unimedizin-mainz.de/neurologie/header/as5f.html for online
calculation of the score and NNS.
ROC-AUC 0.77 for AS5F and 0.68 for age as continuous
variable, p = 0.004).
AS5F performance beyond 72 hours of Holter-
ECG monitoring
There were no major differences in patient characteristics
between the patients diagnosed with pAF within 72 hours and
those diagnosed beyond 72 hours of prolonged Holter-ECG
(for details see table 2). The external validation cohort con-
sists of 2 studies (Find-AF, Find-AFrandomized) investigating
prolonged Holter-ECG beyond 72 hours, namely, 7 days in
Find-AF and 10 days in Find-AFrandomized. Although AS5F
was actually developed to capture the pAF risk within the first
72 hours, we also investigated its performance for longer
monitoring periods. As expected, when considering all pAF
events including those observed beyond 3 days, AS5F
underestimates the true pAF risks (calibration line intercept
0.01 and slope 1.81). AS5F was developed on pAF events
within a 72-hour period; hence, the additional risk of pAF
after 72 hours cannot be accurately reflected by the score.
However, the score is still suitable to differentiate between
low- and high-risk patients as reflected by an ROC-AUC of
0.77, which does not differ from its discriminative perfor-
mance for the 72-hour outcome (p = 0.82; figure 2B).
Moreover, the superiority of AS5F over the CHADS2 score in
discriminating between high- and low-risk patients is pre-
served even for prolonged Holter-ECG monitoring beyond
72 hours (NRI 0.22 with p = 0.009; ROC-AUC 0.77 for
AS5F and 0.63 for CHADS2 score, p < 0.001; figure 2C)
and becomes even more superior in comparison to age alone
as continuous variable for monitoring intervals beyond 72
hours (NRI 0.22 with p = 0.004; ROC-AUC 0.77 for AS5F
and 0.68 for age as continuous variable, p = 0.004).
e120 Neurology | Volume 92, Number 2 | January 8, 2019
Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
Discussion
We were able to develop and validate a simple clinical score to
predict the detection of pAF during prolonged 72-hour
Holter monitoring after acute ischemic stroke. The AS5F
score could be used to identify high-risk patients for pAF, with
the potential to reduce the NNS down to 3. This allows the
clinician to allocate time- and resource-consuming prolonged
Holter monitoring to patients at highest risk when there are
not sufficient resources available for universal prolonged ECG
monitoring. Moreover, our data suggest that AS5F can also be
used to direct further prolongation of ECG monitoring be-
yond 72 hours in patients at high risk of pAF.
The detection of pAF after ischemic stroke during stroke-unit
diagnostic workup remains crucial for the introduction of an
appropriate secondary preventive therapy, as it triggers the
of the most effective secondary stroke prevention therapies.15
However, the methods (e.g., Holter-ECG, stroke-unit te-
lemetry or implantable devices) and the appropriate duration
of monitoring currently remain a matter of intense debate.16
Recent literature supports a timely initiation of monitoring
whereas AF detected at longer intervals after the index stroke
is unlikely to be causative but could still identify high-risk
patients who may develop cardioembolic stroke and therefore
motivate changing secondary preventive medication to anti-
coagulation.17 Currently, there is no doubt that prolonged
total duration of monitoring is one of the key determinants for
increased detection rates of AF.7,17 However, one has to keep
in mind the resource-consuming analysis of increasing ECG
data18 and the reality in clinical routine practice, with less than
20% of patients presenting with ischemic stroke undergoing
ECG monitoring beyond 24 hours.8,9 Our score might allow
clinicians an appropriate patient selection for prolonged
monitoring if limited resource capability necessitates the
prioritization of patients’ diagnostics.
Regarding the clinical use of the score, we must clarify that the
value of 67.5 AS5F points (displayed in data available from
Dryad, figure 3, doi.org/10.5061/dryad.ms3m3n2), which
discriminates between high- and low-risk patients, is not
meant as a clear cutoff and there are rather methodological
reasons to compare the AS5F score to existing scoring sys-
tems. Thus, it is important to note that the NNS and the
probability of AF detection within 72 hours of ECG moni-
toring for individual patients calculated under unimedizin-
mainz.de/neurologie/header/as5f.html should both help
physicians in clinical decision-making and motivate patients
to tolerate prolonged ECG monitoring. ECG monitoring of
72 hours should not be withheld from patients classified in the
low-risk group.
AS5F has a high performance for detection of patients at high
risk of AF after ischemic stroke or TIA within a prolonged
Holter-ECG of 72 hours. An 82-year-old patient after an acute
stroke with NIHSS of 6 points (AS5F score 83) has a 19%
Neurology.org/N
Figure 1 Predicted risk of detection of AF within 72 hours of prolonged Holter-ECG monitoring by AS5F score
probability (NNS 5) of having pAF during 72-hour moni-
toring, whereas the probability for a 46-year-old patient with
transient symptoms (AS5F score 35) lies below 1% (pre-
dicted risk 0.25%, NNS 400; figure 1B). An advantage of the
AS5F score is its external validity featuring a high degree of
generalizability in the Caucasian population, as shown by its
performance on different external validation cohorts from
central Europe. In contrast to recent randomized trials that
limited the detection of pAF with prolonged monitoring
methods to patients with only so-called cryptogenic or strokes
of unknown sources (Event Monitor Belt for Recording Atrial
Fibrillation after a Cerebral Ischemic Event [EMBRACE],19
Cryptogenic Stroke and Underlying AF [CRYSTAL-AF]20),
representing only a minority of patients with stroke, our
model was developed and validated in almost unselected
patients with stroke and allows a wide applicability. In addi-
tion, AS5F performed even better than classic scoring systems
such as the CHADS2 score,12 which is reflected by a signifi-
cant NRI in discriminating between high- and low-risk
patients for pAF detection within prolonged monitoring of
72 hours.
Neurology.org/N
Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
(A) Predicted risk of detection of pAF increases with total
score points using the equation depicted in data available
from Dryad (figure 1, doi.org/10.5061/dryad.ms3m3n2).
(B) Nomogram to determine points for the different score
items (Age, Stroke Severity NIHSS >5 to Find AF), which are
added to get total points that then relate to the predicted
AF risk and NNS. The green and red lines display 2 hypo-
thetical cases. The green line demonstrates an example of
a 46-year-old patient presenting with TIA and therefore an
AS5F score of 36 leading to a predicted AF risk of 0.25%
and NNS of 400. The red line represents an example of an
82-year-old patient presenting with stroke and NIHSS >5
points leading to an AS5F score of 83, a predicted AF risk of
19% and NNS of 5. AS5F = Age, Stroke Severity NIHSS >5 to
Find AF; NIHSS = NIH Stroke Scale; NNS = number needed
to screen; pAF = paroxysmal atrial fibrillation.
The superiority of AS5F might be explained by the fact that
current scoring systems such as CHADS2 (which initially was
developed to predict the risk of stroke reoccurrence in
patients with AF but recently was even proven to predict the
occurrence of AF21) were not developed in patient cohorts
after ischemic stroke, and moreover, the detection of AF was
assessed mostly by routine care and not by prolonged ECG
monitoring with a centralized data analysis. In the score de-
velopment procedure, a linear contribution of age to the linear
predictor was shown to be most appropriate for the logistic
regression model. Hence, age is included with 0.76 total score
points per year into AS5F. This leads to a more appropriate
risk prediction of AF detection with increasing age by AS5F
compared to CHADS2, which only considers whether age is
above 75. As age is one of the most important risk factors for
AF,22 we also compared the performance of AS5F to age alone
as continuous variable to discriminate between high- and
low-risk patients and observed a larger AUC for AS5F. This
discriminative superiority of AS5F over age as continuous
variable becomes statistically apparent for monitoring inter-
vals beyond 72 hours.
Neurology | Volume 92, Number 2 | January 8, 2019 e121
 Figure 2 ROC-AUC for prediction of new-onset pAF based on AS5F

(A) AS5F used for prediction of new-onset pAF within a monitoring interval of 72 hours (AUC 0.752) compared to CHADS2 score (AUC 0.607, p = 0.0032). (B) AS5F
used for prediction of new-onset pAF within a monitoring interval less than 72 hours (AUC 0.75) and longer than 72 hours (AUC 0.77, p = 0.82) revealed no
statistical difference in score performance between the 2 patient cohorts. (C) AS5F used for prediction of new-onset pAF within the monitoring interval of the
study period (Find-AF: 7 days, FIND-AFrandomized: 10 days, AUC 0.765) compared to CHADS2 score (AUC 0.629, p < 0.001). AS5F = Age, Stroke Severity NIHSS >5
to Find AF; AUC = area under the curve; CHADS2 = congestive heart failure, hypertension, age >75, diabetes mellitus, stroke/TIA; NIHSS = NIH Stroke Scale;
NRI = net reclassification improvement; pAF = paroxysmal atrial fibrillation; ROC = receiver operating characteristic curve.

With this in mind, the AS5F score can be used in different
ways. First, to illustrate the high diagnostic effectiveness of
prolonged Holter-ECG monitoring within 72 hours, already
recommended by the 2016 European Society of Cardiology
guidelines3 for management of AF in patients after ischemic
stroke, but is not yet anticipated in the real-world setting.8
With an NNS of 3 for the highest observed AS5F points
within the study population, both patients and physicians will
be motivated to perform prolonged Holter-ECG monitoring.
This will hopefully increase the current low frequencies of
patients (outside of well-controlled clinical trials8,9) un-
dergoing Holter-ECG monitoring beyond 24 hours in clinical
practice. More interestingly, the AS5F score might keep its
discriminative performance also with pAF detection durations
beyond an interval of 72 hours: even during prolonged
monitoring from 3 to 7 or up to 10 days. Despite the limita-
tion that AS5F was developed for monitoring intervals up to
72 hours within the IDEAS cohort, the score might also be
used to select patients to undergo even longer monitoring
times; however, we acknowledge that a validation cohort for
intervals beyond 72 hours is missing and therefore needs
replication in further studies.
There are several published risk scores that aim to predict
detection of AF after ischemic stroke. However, they mostly
include echocardiographic measurements23–28 and laboratory
parameters,23,25,26 which are not available in the majority of
patients during the acute stroke admission and in common
daily practice. Some scores even require further brain
imaging26,28 to detect hemorrhagic transformation or cortical
involvement as a composite item of the score,26 which might
be one reason that these scores have not become everyday
practice. The available scores considering only clinical items
use a composition of Holter-ECG and stroke-unit telemetry
for diagnostic workup21,29,30 or focus on a subset of patients,
such as those after catheter ablation due to atrial flutter.31 In

e122 Neurology | Volume 92, Number 2 | January 8, 2019 Neurology.org/N

Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
addition, there are only limited scores investigating a single
method of ECG recording (e.g., Holter-ECG)23,24 and
a monitoring interval of 72 hours.25 Most use a combination
of conventional stroke-unit telemetry and an additional 24-
hour Holter-ECG for the risk-score development.25,26,28,29
The AS5F score aims to predict pAF, based on solely clinical
parameters. Its strength is the development and validation in
multiple nearly unselected large cohorts of acute ischemic
stroke patients with the same method of prolonged ECG
recording (Holter-ECG) with a centralized structured anal-
ysis. Furthermore, AS5F might even be useful in a patient
cohort beyond the monitoring interval of 72 hours, namely,
up to 10 days of prolonged Holter monitoring (as seen for the
Find-AF11 and Find-AFrandomized7 study cohorts).
It is likely that the prediction accuracy of the AS5F score
might be further improved by adding laboratory parameters
such as brain natriuretic peptide32 or echocardiographic
measurements.33 Recently, De Marchis et al.34 showed mid-
regional proatrial natriuretic peptide to be associated with AF
as cardioembolic stroke etiology. Moreover, the evidence on
echocardiographic measurements is growing; recently, lower
atrial strain was shown to be predictive for AF after ischemic
stroke.35 Unfortunately, these parameters were not collected
in the IDEAS cohort, which is a limitation of the proposed
AS5F score. There are other possible predictors for AF that
we did not include in our analysis because they were not
available in all 3 cohorts (e.g., P terminal force in V1, embolic
stroke pattern on cerebral imaging). However, no pattern of
acute brain infarction was associated in CRYSTAL-AF,36 and
AF detection rate was similar in patients with cryptogenic vs
noncryptogenic stroke in Find-AFrandomized.7 However, to
maximize the global usefulness and easy applicability, we
choose to include only available clinical parameters at
patients’ presentation for score development.
The ongoing Embolic Stroke of Undetermined Source (ESUS)
trials are examining the benefit of oral anticoagulation com-
pared to acetylsalicylic acid in patients with embolic stroke of
unknown source. Because stroke etiology might change from
the index stroke in up to 50% of the patients,37 all stroke
patients have to be considered at high risk of future car-
diovascular events, irrespective of the suspected cause of the
index stroke. Thus, the proposed AS5F score, developed in
an unselected patient cohort, predicts the risk of pAF de-
tection within 72 hours of ECG monitoring in almost all
stroke patients and will be of relevance irrespective of the
results of the ESUS trials.
The major strength of this study is the large sample size of
patients (n = 1,556) with at least 72 hours of prolonged
Holter-ECG monitoring after ischemic stroke, which provides
the ability to develop a robust score for prediction of pAF
within prolonged monitoring times. Moreover, the central-
ized analysis of all patient data in a core laboratory makes
a misclassification of pAF diagnosis unlikely. One limitation
might be that the validity of the prediction beyond 72 hours of
Neurology.org/N
Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
monitoring is only based on the assumption of equally dis-
tributed patient and pAF detection characteristics and
therefore does not share the same prospective power to
predict pAF compared to <72 hours of monitoring. In addi-
tion, the CHADS2 score was not specifically created to predict
pAF and we were not able to assess all variables of the newer
CHA2DS2Vasc score38 from the case report forms. Second,
the predicted risk will probably change when applying the risk
score to patient populations that strongly differ in the corre-
lation between variables included in the risk score and con-
founders, in particular regarding the small sample size of
patients diagnosed with AF. However, a well-reproducible
prediction performance was observed in the validation cohort.
Nevertheless, this might be different in other study cohorts.
With this in mind, we stress that the 3 included studies within
this analysis recruited patients in central Europe and therefore
the applicability to further ethnicities, in which the biological
age and the risk factors of AF might differ,39 has to be vali-
dated in further studies.
In short, AS5F is a risk score based solely on clinical param-
eters that can easily be used by clinicians to select patients for
prolonged Holter-ECG monitoring to increase the diagnostic
yield of pAF after ischemic stroke or TIA and might improve
the secondary preventive strategy in order to prevent re-
current ischemic strokes.
Author contributions
Timo Uphaus: design and conceptualization of the study,
analysis and interpretation of the data, drafting and revision
the manuscript. Mark Weber-Krüger: design and conceptu-
alization of the study, analysis and interpretation of the data,
revision the manuscript. Martin Grond: design and concep-
tualization of the study, revision the manuscript. Gerrit
Toenges: design and conceptualization of the study, analysis
and interpretation of the data, drafting and revision the
manuscript. Antje Jahn-Eimermacher: design and conceptu-
alization of the study, analysis and interpretation of the data,
drafting and revision the manuscript. Marek Jauss: design and
conceptualization of the study, revision the manuscript.
Paulus Kirchhof: design and conceptualization of the study,
revision the manuscript. Rolf Wachter: design and concep-
tualization of the study, analysis and interpretation of the data,
drafting and revision the manuscript. Klaus Gröschel: design
and conceptualization of the study, analysis and interpretation
of the data, drafting and revision the manuscript.
Acknowledgment
The authors thank Dr. L. Rosin for his effort within the IDEAS
trial and Dr. Cheryl Ernest for proofreading the manuscript.
Study funding
No targeted funding reported.
Disclosure
T. Uphaus, M. Weber-Krüger, G. Toenges, A. Jahn-
Eimermacher, and M. Jauss report no disclosures relevant to
Neurology | Volume 92, Number 2 | January 8, 2019 e123
 the manuscript. P. Kirchhof receives research support from
European Union, British Heart Foundation, Leducq Foun-
dation, Medical Research Council (UK), and German Centre
for Cardiovascular Research, from several drug and device
companies active in atrial fibrillation, and has received hon-
oraria from several such companies. Martin Grond received
honoraria for participation in clinical trials, advisory boards or
speakers bureau for Boehringer Ingelheim, Bayer Vital,
Bristol-Myers Squibb, Pfizer, Daiichi-Sankyo, Medtronics,
and Sanofi-Aventis. R. Wachter reports having been an in-
vestigator or consultant for, or received fees from, Bayer,
Berlin-Chemie, Bristol-Myers Squibb, Boehringer Ingelheim,
Boston Scientific, CVRx, Gilead, Johnson & Johnson, Med-
tronic, Novartis, Pfizer, Relypsa, Sanofi, Servier since 2003.
He received research grants from Boehringer Ingelheim, Eu-
ropean Union, and Bundesministerium für Bildung und
Forschung (BMBF). K. Gröschel reports personal fees and
nonfinancial support from Bayer, personal fees and non-
financial support and grant from Boehringer Ingelheim, per-
sonal fees from Bristol-Myers Squibb, personal fees from
Daiichi-Sankyo, personal fees and nonfinancial support from
Pfizer. Go to Neurology.org/N for full disclosures.
Publication history
Received by Neurology February 14, 2018. Accepted in final form August
22, 2018.
References
1. Lin HJ, Wolf PA, Kelly-Hayes M, et al. Stroke severity in atrial fibrillation. The
Framingham Study. Stroke 1996;27:1760–1764.
2. January CT, Wann LS, Alpert JS, et al. 2014 AHA/ACC/HRS guideline for the
management of patients with atrial fibrillation: a report of the American College of
Cardiology/American Heart Association Task Force on practice guidelines and the
Heart Rhythm Society. J Am Coll Cardiol 2014;64:e1–e76.
3. Kirchhof P, Benussi S, Kotecha D, et al. 2016 ESC guidelines for the management of
atrial fibrillation developed in collaboration with EACTS. Eur Heart J 2016;37:
2893–2962.
4. Freedman B, Camm J, Calkins H, et al. Screening for atrial fibrillation: a report of the
AF-SCREEN international collaboration. Circulation 2017;135:1851–1867.
5. Freedman B, Potpara TS, Lip GY. Stroke prevention in atrial fibrillation. Lancet 2016;
388:806–817.
6. Hart RG, Pearce LA, Aguilar MI. Meta-analysis: antithrombotic therapy to prevent
stroke in patients who have nonvalvular atrial fibrillation. Ann Intern Med 2007;146:
857–867.
7. Wachter R, Gröschel K, Gelbrich G, et al; Find AFrandomised Investigators and
Coordinators. Holter-electrocardiogram-monitoring in patients with acute ischaemic
stroke (Find-AFRANDOMISED): an open-label randomised controlled trial. Lancet
Neurol 2017;16:282–290.
8. Rizos T, Quilitzsch A, Busse O, et al. Diagnostic work-up for detection of paroxysmal
atrial fibrillation after acute ischemic stroke: cross-sectional survey on German stroke
units. Stroke 2015;46:1693–1695.
9. Giruparajah M, Bosch J, Vanassche T, et al. Global survey of the diagnostic
evaluation and management of cryptogenic ischemic stroke. Int J Stroke 2015;
10:1031–1036.
10. Grond M, Jauss M, Hamann G, et al. Improved detection of silent atrial fibrillation
using 72-hour Holter ECG in patients with ischemic stroke: a prospective multicenter
cohort study. Stroke 2013;44:3357–3364.
11. Stahrenberg R, Weber-Krüger M, Seegers J, et al. Enhanced detection of paroxysmal
atrial fibrillation by early and prolonged continuous Holter monitoring in patients
with cerebral ischemia presenting in sinus rhythm. Stroke 2010;41:2884–2888.
12. Gage BF, Waterman AD, Shannon W, Boechler M, Rich MW, Radford MJ. Validation
of clinical classification schemes for predicting stroke: results from the National
Registry of Atrial Fibrillation. JAMA 2001;285:2864–2870.
e124 Neurology | Volume 92, Number 2 | January 8, 2019
Copyright ª 2018 American Academy of Neurology. Unauthorized reproduction of this article is prohibited.
13. Pencina MJ, D’Agostino RB Sr, D’Agostino RB Jr, Vasan RS. Evaluating the added
predictive ability of a new marker: from area under the ROC curve to reclassification
and beyond. Stat Med 2008;27:157–172; discussion 207–212.
14. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or
more correlated receiver operating characteristic curves: a nonparametric approach.
Biometrics 1988;44:837–845.
15. Hankey GJ. Stroke. Lancet 2017;389:641–654.
16. Sposato LA, Cipriano LE, Saposnik G, Ruiz Vargas E, Riccio PM, Hachinski V.
Diagnosis of atrial fibrillation after stroke and transient ischaemic attack: a systematic
review and meta-analysis. Lancet Neurol 2015;14:377–387.
17. Kishore A, Vail A, Majid A, et al. Detection of atrial fibrillation after ischemic stroke or
transient ischemic attack: a systematic review and meta-analysis. Stroke 2014;45:
520–526.
18. Uphaus T, Grings A, Gröschel S, et al. Automatic detection of paroxysmal atrial
fibrillation in patients with ischaemic stroke: better than routine diagnostic workup?
Eur J Neurol 2017;7:990–994.
19. Gladstone DJ, Spring M, Dorian P, et al. Atrial fibrillation in patients with cryptogenic
stroke. N Engl J Med 2014;370:2467–2477.
20. Sanna T, Diener HC, Passman RS, et al. Cryptogenic stroke and underlying atrial
fibrillation. N Engl J Med 2014;370:2478–2486.
21. Fauchier L, Clementy N, Pelade C, Collignon C, Nicolle E, Lip GY. Patients with
ischemic stroke and incident atrial fibrillation: a nationwide cohort study. Stroke
2015;46:2432–2437.
22. Schnabel RB, Sullivan LM, Levy D, et al. Development of a risk score for atrial
fibrillation (Framingham Heart Study): a community-based cohort study. Lancet
2009;373:739–745.
23. Fujii S, Shibazaki K, Kimura K, Sakai K, Aoki J. A simple score for predicting par-
oxysmal atrial fibrillation in acute ischemic stroke. J Neurol Sci 2013;328:83–86.
24. Figueiredo MM, Rodrigues AC, Alves MB, Neto MC, Silva GS. Score for atrial
fibrillation detection in acute stroke and transient ischemic attack patients in a Bra-
zilian population: the acute stroke atrial fibrillation scoring system. Clinics 2014;69:
241–246.
25. Yoshioka K, Watanabe K, Zeniya S, et al. A score for predicting paroxysmal atrial
fibrillation in acute stroke patients: iPAB score. J Stroke Cerebrovasc Dis 2015;24:
2263–2269.
26. Seo WK, Kang SH, Jung JM, Choi JY, Oh K. Novel composite score to predict atrial
fibrillation in acute stroke patients: AF predicting score in acute stroke. Int J Cardiol
2016;209:184–189.
27. Bugnicourt JM, Flament M, Guillaumont MP, et al. Predictors of newly diagnosed
atrial fibrillation in cryptogenic stroke: a cohort study. Eur J Neurol 2013;20:
1352–1359.
28. Suissa L, Mahagne MH, Lachaud S. Score for the targeting of atrial fibrillation: a new
approach to diagnosing paroxysmal atrial fibrillation. Cerebrovasc Dis 2011;31:
442–447.
29. Giralt-Steinhauer E, Cuadrado-Godia E, Soriano-Tárraga C, et al. New-onset parox-
ysmal atrial fibrillation diagnosis in ischemic stroke patients. Eur Neurol 2015;74:
211–217.
30. Baturova MA, Lindgren A, Carlson J, et al. Predictors of new onset atrial fibrillation
during 10-year follow-up after first-ever ischemic stroke. Int J Cardiol 2015;199:
248–252.
31. Chen K, Bai R, Deng W, et al. HATCH score in the prediction of new-onset atrial
fibrillation after catheter ablation of typical atrial flutter. Heart Rhythm 2015;12:
1483–1489.
32. Seegers J, Zabel M, Gruter T, et al. Natriuretic peptides for the detection of parox-
ysmal atrial fibrillation. Open Heart 2015;2:e000182.
33. Stahrenberg R, Edelmann F, Haase B, et al. Transthoracic echocardiography to rule
out paroxysmal atrial fibrillation as a cause of stroke or transient ischemic attack.
Stroke 2011;42:3643–3645.
34. De Marchis GM, Schneider J, Weck A, et al. Midregional proatrial natriuretic peptide
improves risk stratification after ischemic stroke. Neurology 2018;90:e455–e465.
35. Pathan F, Sivaraj E, Negishi K, et al. Use of atrial strain to predict atrial fibrillation after
cerebral ischemia. JACC Cardiovasc Imaging 2018;11:1557–1565.
36. Bernstein RA, Di Lazzaro V, Rymer MM, et al. Infarct topography and detection of
atrial fibrillation in cryptogenic stroke: results from CRYSTAL AF. Cerebrovasc Dis
2015;40:91–96.
37. Hillen T, Coshall C, Tilling K, et al. Cause of stroke recurrence is multifactorial:
patterns, risk factors, and outcomes of stroke recurrence in the South London Stroke
Register. Stroke 2003;34:1457–1463.
38. Lip GY, Nieuwlaat R, Pisters R, Lane DA, Crijns HJ. Refining clinical risk stratification
for predicting stroke and thromboembolism in atrial fibrillation using a novel risk
factor-based approach: the Euro Heart Survey on Atrial Fibrillation. Chest 2010;137:
263–272.
39. Oldgren J, Healey JS, Ezekowitz M, et al. Variations in cause and management of atrial
fibrillation in a prospective registry of 15,400 emergency department patients in 46
countries: the RE-LY Atrial Fibrillation Registry. Circulation 2014;129:1568–1576.
Neurology.org/N
 Development and validation of a score to detect paroxysmal atrial fibrillation after
stroke
Timo Uphaus, Mark Weber-Krüger, Martin Grond, et al.
Neurology 2019;92;e115-e124 Published Online before print December 7, 2018
DOI 10.1212/WNL.0000000000006727

This information is current as of December 7, 2018

Updated Information & including high resolution figures, can be found at:
Services http://n.neurology.org/content/92/2/e115.full

References This article cites 39 articles, 14 of which you can access for free at:
http://n.neurology.org/content/92/2/e115.full#ref-list-1
Subspecialty Collections This article, along with others on similar topics, appears in the
following collection(s):
All Cerebrovascular disease/Stroke
http://n.neurology.org/cgi/collection/all_cerebrovascular_disease_strok
e
All Clinical Neurology
http://n.neurology.org/cgi/collection/all_clinical_neurology
Cardiac
http://n.neurology.org/cgi/collection/cardiac
Infarction
http://n.neurology.org/cgi/collection/infarction
Stroke prevention
http://n.neurology.org/cgi/collection/stroke_prevention
Permissions & Licensing Information about reproducing this article in parts (figures,tables) or in
its entirety can be found online at:
http://www.neurology.org/about/about_the_journal#permissions
Reprints Information about ordering reprints can be found online:
http://n.neurology.org/subscribers/advertise

Neurology ® is the official journal of the American Academy of Neurology. Published continuously since
1951, it is now a weekly with 48 issues per year. Copyright © 2018 American Academy of Neurology. All
rights reserved. Print ISSN: 0028-3878. Online ISSN: 1526-632X.