Comparative assessment of published atri

(1)

Comparative assessment of published atrial

_ﬁ

brillation stroke risk strati

_ﬁ

cation

schemes for predicting stroke, in a non-atrial

_ﬁ

brillation population: The Chin-Shan

Community Cohort Study

Gregory Y.H. Lip

a,

⁎

, Hung-Ju Lin

b

, Kuo-Liong Chien

a,b,c,

⁎⁎

, Hsiu-Ching Hsu

c

, Ta-Chen Su

c

,

Ming-Fong Chen

c

_{, Yuan-Teh Lee}

c,d

a

University of Birmingham Centre for Cardiovascular Sciences, City Hospital, Birmingham, B18 7QH, United Kingdom b

Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan c

Institute of Epidemiology & Preventive Medicine, College of Public Health, National Taiwan University, Taipei, Taiwan d_{Institute of Clinical Medical Science, Chinese Medical University, Taichung, Taiwan}

a b s t r a c t

a r t i c l e

i n f o

Article history:

Received 2 August 2012

Received in revised form 15 September 2012 Accepted 22 September 2012

Available online 13 October 2012

Keywords:

Stroke risk stratiﬁcation Atrialﬁbrillation CHADS2

Background:In patients at high risk of stroke, such as atrialﬁbrillation (AF), there has been great interest in developing stroke risk prediction schemes for identifying those at high risk of stroke. Stroke risk prediction schemes have also been developed in non-AF populations, but are limited by lack of simplicity, which is more evident in schemes used in AF populations. We hypothesized that contemporary stroke risk stratiﬁ ca-tion schemes used in assessing AF patients could predict stroke and thromboembolism in a non-AF commu-nity population, comparably to that seen in AF populations.

Methods:We tested the CHADS2and CHA2DS2-VASc schemes, as well as the AF stroke risk stratiﬁcation

schemes from the Framingham study, Rietbrock et al., 2006 ACC/AHA/ESC guidelines, the 8th American Col-lege of Cardiology (ACCP) guidelines and NICE, for predicting stroke in a large community cohort of non-AF subjects, the Chin-Shan Community Cohort Study.

Results:The tested schemes had variable classiﬁcation into low, moderate and high risk strata, with the pro-portion classiﬁed as low risk ranging from 5.4% (Rietbrock et al. to 59.0% (CHADS2classical). Rates of stroke

also varied in those classiﬁed as‘low risk’ranging from 1.1% (Rietbrock et al. to 3.5% (Framingham). All com-mon risk schemes had broadly similar c-statistics, ranging from 0.658 (Framingham) to 0.728 (CHADS2

clas-sical) when assessed as a continuous risk variable for predicting stroke in this population, with clear overlap between the 95% CIs. In an exploratory analysis amongst AF subjects in our population, the c-statistics were broadly similar to those seen in non-AF subjects.

Conclusion:Contemporary stroke risk stratiﬁcation schema used for AF can also be applied to non-AF populations with a similar (modest) predictive value. Given their simplicity (e.g. CHADS2score), these scores

could potentially be used for a‘quick’evaluation of stroke risk in non-AF populations, in a similar manner to AF populations.

1. Introduction

In patients at high risk of stroke, such as atrial

ﬁ

brillation (AF),

there has been great interest in developing stroke risk prediction

schemes for identifying those at high risk of stroke. Stroke risk

predic-tion schemes have also been developed in non-AF populapredic-tions, but

are limited by lack of simplicity

[1]

, which is more evident in schemes

used in AF populations, such as the CHADS

2

and CHA

2

DS

2

-VASc

schemes

[2,3]

.

The CHADS

2

scheme is an amalgamation of stroke risk factors

identi

_ﬁ

ed from 2 trial-based stroke risk strati

_ﬁ

cation schemes, the

AF Investigators and the SPAF-1 schemes

[2]

. However, the CHADS

2

scheme has many limitations, and does not include many stroke risk

factors

[4,5]

. To complement the CHADS

2

scheme, the CHA

2

DS

2

-VASc

has been developed

[3]

, by being more inclusive (rather than

exclu-sive) of stroke risk factors. The CHA

2

DS

2

-VASc scheme has been

shown to be as good as (and possibly better) than the CHADS

2

scheme in predicting high risk patients with AF who develop stroke

and thromboembolism (TE)

[6,7]

, but performs particularly well in

identifying those patients with AF who are

‘

truly low risk

’

of

throm-boembolism, who do not need any antithrombotic therapy

[6

_–

10]

.

–

⁎ Corresponding author. Tel.: +44 121 5075080; fax: +44 121 5544083.

⁎⁎ Corresponding author at: Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan. Tel.: +886 2 23123456x62830; fax: +886 2 23511955.

E-mail addresses:g.y.h.lip@bham.ac.uk(G.Y.H. Lip),klchien@ntu.edu.tw

(K.-L. Chien).

http://dx.doi.org/10.1016/j.ijcard.2012.09.148

Contents lists available at

ScienceDirect

International Journal of Cardiology

(2)

The CHA

2

DS

2

-VASc scheme is used in the 2010 European Society of

Cardiology guidelines on AF management

[11]

.

The CHADS

2

score has been applied to small cohorts of non-AF

populations, and has been reported to have modest predictive value

for predicting stroke and thromboembolism

[2,12]

. One recent

analy-sis found the CHADS

2

score to have predictive value for adverse

car-diovascular events in patients admitted with stable coronary artery

disease

[13]

. In the REACH registry, the CHADS

2

score was also related

to cardiovascular events in a large population of patients with

atherothrombosis

[14]

. However, we are unaware of any

comprehen-sive analysis of contemporary AF stroke risk strati

ﬁ

cation schemes in

a large prospective community cohort of non-AF subjects, let alone in

a Far Eastern population.

We therefore hypothesized that contemporary stroke risk strati

ﬁ

ca-tion schemes could predict stroke and thromboembolism, comparably

to that seen in AF populations. To test this hypothesis we applied the

CHADS

2

and CHA

2

DS

2

-VASc schemes, as well as the AF stroke risk

strat-i

ﬁ

cation schemes from the Framingham study

[15]

, Rietbrock et al.

[16]

,

2006 ACC/AHA/ESC guidelines

[17]

, the 8th American College of

Cardiol-ogy (ACCP) guidelines

[18]

and NICE

[19]

, to a large community cohort

of non-AF subjects, the Chin-Shan Community Cohort Study.

2. Methods

2.1. Study design and study participants

Details of this cohort study have been published previously[20]. In brief, the Chin-Shan Community Cohort (CCCC) Study began in 1990 by recruiting 1703 men and 1899 women of Chinese ethnicity aged >35 years from the town of Chin-Shan, 30 km north of metropolitan Taipei, Taiwan. Information about lifestyle and medical conditions and anthropometric measures was assessed by interview questionnaires and physical examinations in 2-year cycles for the initial 6 years; the validity and reli-ability of the collected data and measurements have been reported in details elsewhere

[20,21]. The cohort was followed up from 1990 to the end of 2007 (a total of 49 281 person-years, median 15.9 years, interquartile range: 12.8 to 16.9 years)[20].

2.2. Description of stroke risk stratiﬁcation schema

The various stroke risk schema compared and/or validated in this‘real world’cohort are summarized inTable 1. The Framingham, CHADS2and CHA2DS2-VASc schemes are

point-based scores, with the Framingham one based on a mathematical formula[15]

and the CHADS based on 1 point for CHAD (congestive heart failure, hypertension, age>75 and diabetes) and 2 points for stroke/TIA[2]. The CHA2DS2-VASc score is based

on 2 points for stroke/TIA and age≥75, and 1 point for CHAD, age 65–75, vascular disease and female gender[3].

In order to compare their predictive ability with other schema for distinguishing low, intermediate and high risk strata, we categorized also the scores into three groups. We deﬁned the CHADS2score in two ways: (i)classical, whereby scores of 0 = low, 1–2 =

intermediate, >2=high risk; or (ii)revised, whereby scores of 0=low, 1=intermediate,

≥2=high risk. The CHA2DS2-VASc score was categorized as 0=low, 1=intermediate

and≥1 as high risk. We categorized the Framingham score in a similar manner to that pro-posed by Fang et al.[22], as follows: score 0–7=low, 8–15=intermediate, 16–31=high risk. In addition to these categorized deﬁnitions (commonly used in clinical practice), the Framingham, CHADS2and CHA2DS2-VASc scores were also tested as continuous variables.

2.3. Follow-up strategy and outcome ascertainment

Procedures for our documentation of incident stroke have been previously described and validated[23–25].

Incident stroke cases were ascertained according to the following standard criteria: a sudden neurological symptom of vascular origin that lasted >24 h with supporting evi-dence from brain imaging studies. Fatal stroke cases were included. Deaths were

identi-fied from official certificate documents and verified by house-to-house visits. The cases were confirmed by cardiologists and neurologists. Transient ischemic attacks were not included in this study, especially since this is a‘soft’endpoint. The National Taiwan University Hospital Committee Review Board approved the study protocol.

2.4. Statistical analysis

We used descriptive analyses with proportions and means (±standard deviation) to describe the validation cohort, categorization of the three risk groups per schema and the event rates per risk group. We calculated the 95% conﬁdence interval of event rates using the binomial approximation. We performed logistic regression with each schema, containing three risk groups, as independent variable and TE during 1 year as dependent variable. We calculated the area under the curve for the receiver-operating characteristic (ROC) which represents the ability of a schema to correctly classify risk for TE events, which is also referred to as the c-statistic (Harrell's c).

The cohort largely consisted of non-AF subjects (n = 3524), but as a sensitivity

exploratoryanalysis, we calculated the c-statistics in a small separate cohort of AF cases (n = 38) within our cohort.

All statistical tests were 2-sided with a Type I error of 0.05, and probability values of

b0.05 were considered statistically signiﬁcant. Analyses were performed with SAS Version 9.1 (SAS Institute, Cary, NC), Stata Version 9.1 (Stata Corporation, College Station, Texas), and R Version 2.9.0 (The R Foundation for Statistical Computing).

3. Results

Our study population and associated risk factors are shown in

Table 2

. As compared to AF subjects, non-AF subjects tended to be

younger, and to have less proportion with use of cigarettes,

hyperten-sion, type 2 diabetes mellitus, heart failure and coronary artery

disease (p

b

0.05). As to prior stroke, there was no difference between

those with or without presence of baseline AF.

Risk strati

ﬁ

cation, incidence of stroke, and predictive ability for

risk stratum amongst the 3524 CCCC study participants without

baseline AF are shown in

Table 3

.

The schemes had variable classi

ﬁ

cation into low, moderate and

high risk strata, with the proportion classi

ﬁ

ed as low risk ranging

from 5.4% [

16 ] to 59.0% (CHADS

2

classical). Rates of stroke also varied

Table 1

Risk stratiﬁcation schemes used to predict thromboembolism in atrialﬁbrillation.

Risk scheme Ref Low risk Intermediate risk High risk

CHADS2(2001)—classical [2] Score 0 Score 1–2 Score 3–6

CHADS2—revised [3] Score 0 Score 1 Score 2–6

Framingham (2003) [24] Score 0–7 Score 8–15 Score 16–31 Rietbrock et al (2008) [16] Score 0 Score 1–5 Score 6–14 NICE guidelines (2006) [19] Ageb65 years with

no moderate/high risk factors

Age≥65 years with no high risk factors Ageb75 years with hypertension, diabetes or vascular diseasea

Previous stroke/TIA or thromboembolic event

Age≥75 years with hypertension, diabetes or vascular disease Clinical evidence of valve disease or heart failure, or impaired left ventricular function

ACC/AHA/ESC guidelines (2006) [17] No risk factors Age≥75 years, or hypertension, or heart failure, or LVEF≤35%, or diabetes

Previous stroke, TIA or embolism, or≥2 moderate risk factors of (age≥75 years, hypertension, heart failure, LVEF≤35%, diabetes)

8th ACCP guidelines (2008) [18] No risk factors Age > 75y, or hypertension, or moderately or severely impaired LVEF and/or heart failure, or diabetes

Previous stroke, TIA or embolism, or≥2 moderate risk factors of (age≥75 years, hypertension, moderately or severely impaired LVEF and/or heart failure, diabetes) CHA2DS2-VASc (2009) [3] No risk factors One‘clinically relevant nonmajor’risk factor

(heart failure/LVEF≤40, hypertension, diabetes, vascular diseasea_{, female gender,}

age 65–74)

Previous stroke, TIA or embolism, or age≥75 years, or≥2

‘clinically

relevant nonmajor’risk factors (heart failure/LVEF≤40, hypertension, diabetes, vascular diseasea_{, female gender,}

age 65–74)

a

(3)

in those classi

ﬁ

ed as

‘

low risk

’

ranging from 1.1% [

16 ] to 3.5%

(Framingham).

All common risk schemes had broadly similar c-statistics, ranging

from 0.658 (Framingham) to 0.728 (CHADS

2

classical) when assessed

as a continuous risk variable for predicting stroke in this population,

with clear overlap between the 95% CIs (

Fig. 1, Table 3

). When analysed

as 3 categories, the c-statistics for modi

ﬁ

ed CHADS

2

[

16 ] and CHA

2

DS

2

-VASc schemes were less impressive compared to other schemes.

In an exploratory analysis amongst the small cohort of AF subjects

in our population (who had 12 stroke events), the point estimates of

the c-statistics were broadly similar for CHA

2

DS

2

-VASc (0.623),

Fra-mingham (0.643), Rietbrock et al. (0.660), CHADS

2

classical (0.595),

CHADS

2

revised (0.595), ACC/AHA/ESC and ACCP (both 0.593), with

much overlap in 95% CIs given the small numbers in this cohort

(full data not shown).

4. Discussion

In this analysis, we show that many contemporary stroke risk

strati

ﬁ

cation schema used for AF can also be applied to non-AF

populations with a similar (modest) predictive value, as re

ﬂ

ected by

the c-statistic. This would have advantages, since schemes such as

CHADS

2

are simple and easily remembered, whilst other stroke risk

scores are usually based on weighted formulae derived from

multi-variate analyses. Of note, the c-statistics in our community study

were broadly similar for the schemes whether applied to non-AF

and AF populations.

The present analysis shows that the CHADS

2

scheme allows a

sim-ple and rapid assessment of stroke risk, even in a non-AF population.

This would enhance rapid clinical assessment of patients who may be

at risk of stroke. Other stroke risk scoring systems have been evident

for many years, and these prediction models for the risk of stroke

have been helpful to guide screening and interventions and to predict

stroke events, but derivation of some prediction models were based

on hospital-based patients with various co-morbid conditions. One

older stroke risk assessment model based on a community cohort,

the Framingham risk score included variables such as age, systolic

blood pressure, antihypertensive therapy, diabetes mellitus, smoking,

history of CVD, AF and LVH

[1]

.

Based on the present Chin-Shan Community Cohort Study, we

re-cently published a model for predicting the 15-year incidence of

stroke in a community-based Chinese adult population, based on

age (8 points), gender (1 point), systolic blood pressure (3 points),

di-astolic blood pressure (2 points), family history of stroke (1 point),

atrial

ﬁ

brillation (3 points), and diabetes (1 point), where the

c-statistic was 0.772 (95% CI, 0.744 to 0.799)

[20,21]

. This model

in-cludes some variables of the CHADS

2

score, but is weighted (and

thus, more complex compared to (say) the CHADS

2

score), includes

family history of stroke and AF as additional variables, and is designed

for

‘

general

’

population assessments.

Nonetheless, the CHADS

2

score has been shown to predict ischemic

stroke in the absence of AF amongst subjects with stable coronary heart

disease

[13]

. Indeed, Welles et al.

[13]

studied 916 non-anticoagulated

Table 2

Baseline characteristics of the study cohort by absence or presence of atrialﬁbrillation (AF)a

Age, years, mean (SD) 54.8 (12.3) 67.0 (10.2) 54.9 (12.3) Age 65–75, % 15.9 36.8 16.1

Age >=75, % 6.4 23.7 6.5

Female, % 53.0 36.8 52.8

Systolic blood pressure, mm Hg, mean(SD) 125 (21) 138 (29) 126 (21) Diastolic blood pressure, mm Hg, mean(SD) 77 (11) 81 (16) 77 (11) Body mass index, kg/m2

, mean (SD) 23.5 (3.4) 24.0 (3.6) 23.5 (3.4) LV ejection fraction,%, mean (SD) 68.4 (10.4) 62.1 (16.1) 68.3 (10.4)

Baseline medical conditions

Alcohol use, % 29.7 39.5 29.9 Current smoker, % 36.1 52.6 36.4 Hypertension, % 29.8 52.6 30.0

Diabetes, % 13.0 31.6 13.3

Heart failure, % 1.4 7.4 1.4 Coronary artery disease, % 3.4 2.6 3.4

Prior stroke, % 2.5 2.6 2.5

a

Only 2621 cases with echocardiography data were available.

Table 3

Risk stratiﬁcation, incidence of stroke, and predictive ability for risk stratum amongst the CCCC study participants without baseline AF status (n = 3524).

Categorization of stroke risk

CHA2DS2-VASc, 0.698 0.658 0.738 0.575 0.551 0.598

% in risk category 23.5 76.2 0.4 Stroke event, n(%) 20 (2.4%) 173 (6.5%) 3 (23.1%)

Framingham 0.658 0.615 0.700 0.648 0.611 0.685

% in risk category 71.5 26.0 2.5 TE events, n(%) 89 3.5 85 9.3 22 25.0

Rietbrock et al. [16] 0.675 0.633 0.717 0.583 0.557 0.608

CHADS2classical 0.728 0.691 0.764 0.709 0.675 0.743

CHADS2revised 0.728 0.691 0.764 0.724 0.688 0.760

ACC/AHA/ESC 2006 0.712 0.676 0.747 0.726 0.690 0.762

ACCP 2008 0.712 0.676 0.747 0.726 0.690 0.762

NICE 2006 Not available 0.720 0.688 0.752

(4)

outpatients with stable coronary heart disease and after 5821

person-years of follow up, the ischemic stroke/TIA was 0.69/100 person-person-years,

and the c-statistic was 0.65. When compared to low risk (CHADS

2

0 –

1)

subjects, the risk of stroke in intermediate risk patients was increased

2.4-fold, and for high risk patients, 4.0-fold. In the present analysis, we

have extended our previous work

[20]

and that of Welles et al.

[13]

to

show that all the stroke risk scores used in AF can also be applied to

non-AF populations with a similar (modest) predictive value to AF

co-horts, even in a Chinese community cohort.

Other risk assessment schemes have concentrated on prediction of

overall cardiovascular risk, with an endpoint that includes

(myocardi-al infarction, coronary heart disease, stroke, and transient ischaemic

attack), rather than the prediction of stroke per se. In a recent analysis

from the United Kingdom, the QRISK cardiovascular disease risk

equa-tion offered an improvement over the Framingham score in identifying

a high risk population for cardiovascular disease in

[26,27]

. The QRISK

score did underestimate the 10 year cardiovascular disease risk, but

the magnitude of underprediction was smaller than the overprediction

with Framingham score. Other models for cardiovascular disease risk

prediction have been described, including ASSIGN

[28]

. Of note, even a

cardiovascular risk prediction score has prognostic implications in

post-stroke patients

[29]

, although some debate over the applicability

of various scores to different ethnic groups is evident

[30]

.

Notwithstanding how the different scores are presented in

Table 1

,

the arti

ﬁ

cial categorization into low, moderate and high risk strata is

perhaps less relevant in the non-AF population, as stroke risk is a

a)

continuous variables

0.00

0.25

0.50

0.75

1.00

Sensitivity

0.00 0.25 0.50 0.75 1.00

1-Specificity

s1 ROC area: 0.6979

s2 ROC area: 0.6575

s3 ROC area: 0.6751

s4 ROC area: 0.7277

s5 ROC area: 0.7277

s6 ROC area: 0.7118

s7 ROC area: 0.7118

Reference

b)

categorical (ie. low/moderate/high) variables

r1 ROC area: 0.5747

r2 ROC area: 0.648

r3 ROC area: 0.5825

r4 ROC area: 0.7092

r5 ROC area: 0.724

r6 ROC area: 0.726

r7 ROC area: 0.726

r8 ROC area: 0.7199

Reference

0.00 0.25 0.50 0.75 1.00

1-Specificity

0.00

0.25

0.50

0.75

1.00

Sensitivity

(5)

continuum

—

in AF, the identi

ﬁ

cation of the

‘

high risk

’

stratum was such

that such patients could be targeted for the

‘

inconvenient

’

oral

anticoagulation available, which was warfarin. This necessity is less

ap-parently with the availability of new oral anticoagulants that overcome

the dis-utility of warfarin, and may also be relatively safer. Indeed, a

Markov decision analysis model recently suggested that anticoagulation

with one of these new

‘

safer

’

agents should even be considered at a

lowered stroke threshold of 0.9%/year amongst AF populations

[31]

.

The c-statistics in the present study for the various schemes were

broadly comparable to the c-statistics derived for the different

schemes in AF patients from the EuroHeart survey

[3]

. Of note, the

c-statistics in our cohort were also broadly comparable to the

explor-atory analysis in the small number of patients with AF in our wider

non-AF study cohort. Nonetheless, it would be dif

ﬁ

cult to make

com-parisons between c-statistics tested in one validation cohort, with

those derived from another one.

4.1. Limitations

This study is limited by its registry-based design, but its strength is

the prospective follow up

[20,21]

. Unfortunately, we only had small

numbers of subjects with AF at baseline, but our limited exploratory

analysis suggests that the risk schemes had broadly similar predictive

value both AF and non-AF subjects. Our AF diagnosis was based on

documented AF, and more intense/prolonged ECG monitoring may

have pick up AF in what was presumed to be non-AF subjects

[32]

.

In-deed, continuous monitoring may identify AF in 30% of patients with

stroke risk factors, but without previous known AF or stroke/TIA over

a mean followup of 1.1 years

[33]

. The presence analysis would

sup-port the possible use of the AF stroke risk strati

ﬁ

cation schema in

non-AF populations. Also, we included all types of stroke in this

study and did not specify ischemic and haemorrhage subtypes. Also,

our cohort would relate to

‘

all stroke

’

as not all patients had detailed

cerebral imaging, but stroke would be a

_‘

hard

_’

endpoint, in contrast to

TIAs (which were not included) which are a

‘

soft

’

endpoint. Of note,

the healthcare system for stroke in this community-based cohort

was consistent over time, and we ascertained the stroke cases

according to careful medical history and hospitalization records.

Fur-ther validation studies of these scores in the general population, as

well as other non-AF populations should be performed, that include

both Asian and non-Asian cohorts.

In conclusion, contemporary stroke risk strati

ﬁ

cation schema used

for AF can also be applied to non-AF populations with a similar

(mod-est) predictive value. Given their simplicity (e.g. CHADS

2

score) and

pending further validation studies, these scores could possibly be

used for a

‘

quick

’

evaluation of stroke risk in non-AF populations, in

a similar manner to AF populations.

Competing interests

Prof Lip has served as a consultant for Bayer, Astellas, Merck, Sano

ﬁ

,

BMS/P

ﬁ

zer, Daiichi-Sankyo, Biotronik, Portola and Boehringer Ingelheim

and has been on the speakers' bureau for Bayer, BMS/P

_ﬁ

zer, Boehringer

Ingelheim, and Sano

ﬁ

Aventis.

Other authors

—

none declared, as relevant to this mauscript.

Acknowledgements

The authors of this manuscript have veri

_ﬁ

ed that they comply

with the principles of ethical publishing in the International Journal

of Cardiology.

References

[1] Wolf PA, D'Agostino RB, Belanger AJ, Kannel WB. Probability of stroke: a risk

pro-ﬁle from the Framingham Study. Stroke 1991;22:312-8.

[2] Gage BF, Waterman AD, Shannon W, Boechler M, Rich MW, Radford MJ. Valida-tion of clinical classiﬁcation schemes for predicting stroke: results from the Na-tional Registry of Atrial Fibrillation. JAMA 2001;285:2864-70.

[3] Lip GY, Nieuwlaat R, Pisters R, Lane DA, Crijns HJ. Refining clinical risk stratifi ca-tion for predicting stroke and thromboembolism in atrialfibrillation using a novel risk factor-based approach: the euro heart survey on atrialfibrillation. Chest 2010;137:263-72.

[4] Karthikeyan G, Eikelboom JW. The CHADS2score for stroke risk stratiﬁcation in

atrialﬁbrillation—friend or foe? Thromb Haemost 2010;104:45-8.

[5] Keogh C, Wallace E, Dillon C, Dimitrov BD, Fahey T. Validation of the CHADS2

clin-ical prediction rule to predict ischaemic stroke. A systematic review and meta-analysis. Thromb Haemost 2011;106:528-38.

[6] Olesen J, Lip GYH, Hansen ML, et al. Validation of risk stratiﬁcation schemes for predicting stroke and thromboembolism in patients with atrialﬁbrillation: a nation-wide cohort study. BMJ 2011 Jan 31;342:d124.

[7] Boriani G, Botto G, Padeletti L, et al. Improving stroke risk stratiﬁcation using the CHADS2and CHA2DS2-VASc risk scores in paroxysmal atrialﬁbrillation patients

by continuous arrhythmia burden monitoring. Stroke 2011;42:1768-70. [8] Van Staa TP, Setakis E, Di Tanna GL, Lane DA, Lip GY. A comparison of risk

strat-iﬁcation schema for stroke in 79884 atrialﬁbrillation patients in general prac-tice. J Thromb Haemost 2011;9:39-48.

[9] Abu-Assi E, Otero-Raviña F, Allut Vidal G, et al. Comparison of the reliability and validity of four contemporary risk stratiﬁcation schemes to predict thromboem-bolism in non-anticoagulated patients with atrial ﬁbrillation. Int J Cardiol 2013;166:205-9.

[10] Chao TF, Lin YJ, Tsao HM, et al. CHADS(2) and CHA(2)DS(2)-VASc scores in the prediction of clinical outcomes in patients with atrialﬁbrillation after catheter ab-lation. J Am Coll Cardiol 2011;58:2380-5.

[11] Camm AJ, Kirchhof P, Lip GY, et al. Guidelines for the management of atrialﬁbrillation: the task force for the management of atrialﬁbrillation of the European Society of Car-diology (ESC). Eur Heart J 2010;31:2369-429.

[12] Lip GY. Stroke in atrialﬁbrillation: epidemiology and thromboprophylaxis. J Thromb Haemost 2011;9(Suppl. 1):344-51.

[13] Welles CC, Whooley MA, Na B, Ganz P, Schiller NB, Turakhia MP. The CHADS2score

predicts ischemic stroke in the absence of atrialﬁbrillation among subjects with cor-onary heart disease: data from the Heart and Soul Study. Am Heart J 2011;162: 555-61.

[14] Goto S, Bhatt DL, Röther J, et al. Prevalence, clinical proﬁle, and cardiovascular out-comes of atrialﬁbrillation patients with atherothrombosis. Am Heart J 2008;156: 855-63 [863.e2].

[15] Wang TJ, Massaro JM, Levy D, et al. A risk score for predicting stroke or death in individuals with new-onset atrialﬁbrillation in the community: the Framingham Heart Study. JAMA 2003;290:1049-56.

[16] Rietbrock S, Heeley E, Plumb J, van Staa T. Chronic atrialﬁbrillation: incidence, prevalence, and prediction of stroke using the Congestive heart failure, Hyperten-sion, Age >75, Diabetes mellitus, and prior Stroke or transient ischemic attack (CHADS2) risk stratiﬁcation scheme. Am Heart J 2008;156:57-64.

[17] Fuster V, Ryden LE, Cannom DS, et al. ACC/AHA/ESC 2006 guidelines for the man-agement of patients with atrialﬁbrillation—executive summary: a report of the American College of Cardiology/American Heart Association Task Force on Prac-tice Guidelines and the European Society of Cardiology Committee for PracPrac-tice Guidelines (Writing Committee to Revise the 2001 Guidelines for the Manage-ment of Patients with Atrial Fibrillation). Eur Heart J 2006;27:1979-2030. [18] Singer DE, Albers GW, Dalen JE, et al. Antithrombotic therapy in atrialﬁbrillation:

American College of Chest Physicians Evidence-Based Clinical Practice Guidelines (8th edition). Chest 2008;133(6 Suppl.):546S-92S.

[19] National Collaborating Centre for Chronic Conditions. Atrialﬁbrillation: national clinical guideline for management in primary and secondary care. London: Royal College of Physicians; 2006.

[20] Chien KL, Su TC, Hsu HC, et al. Constructing the prediction model for the risk of stroke in a Chinese population: report from a cohort study in Taiwan. Stroke 2010;41:1858-64.

[21] Lee YT, Lin RS, Sung FC, et al. Chin-Shan Community Cardiovascular Cohort in Taiwan: baseline data andﬁve-year follow-up morbidity and mortality. J Clin Epidemiol 2000;53:836-46.

[22] Fang MC, Go AS, Chang Y, et al. Comparison of risk stratiﬁcation schemes to pre-dict thromboembolism in people with non-valvularatrialﬁbrillation. J Am Coll Cardiol 2008;51:810-5.

[23] Chien KL, Su TC, Jeng JS, et al. Carotid artery intima-media thickness, carotid plaque and coronary heart disease and stroke in Chinese. PLoS One 2008;3: e3435.

[24] Chien KL, Sung FC, Hsu HC, Su TC, Lin RS, Lee YT. Apolipoprotein A1 & B, and stroke events in a community-based cohort in Taiwan: report of Chin-Shan Community Cardiovascular Study. Stroke 2002;33:39-44.

[25] Chen PC, Chien KL, Hsu HC, Su TC, Sung FC, Lee YT. Metabolic syndrome and C-reactive protein in stroke prediction: a prospective study in Taiwan. Metabo-lism 2009;58:772-8.

[26] Collins GS, Altman DG. An independent external validation and evaluation of QRISK cardiovascular risk prediction: a prospective open cohort study. BMJ 2009;339:b2584.

[27] Hippisley-Cox J, Coupland C, Vinogradova Y, et al. Predicting cardiovascular risk in England and Wales: prospective derivation and validation of QRISK2. BMJ 2008;336: 1475-82.

(6)

[29] Ovbiagele B, Liebeskind DS, Kim D, Ali LK, Pineda S, Saver JL. Prognostic value of Framingham Cardiovascular Risk Score in hospitalized stroke patients. J Stroke Cerebrovasc Dis 2011;20:222-6.

[30] Bhopal R, Fischbacher C, Vartiainen E, Unwin N, White M, Alberti G. Predicted and observed cardiovascular disease in South Asians: application of FINRISK, Framing-ham and SCORE models to Newcastle Heart Project data. J Public Health (Oxf) 2005;27:93–100.

[31] Eckman MH, Singer DE, Rosand J, Greenberg SM. Moving the tipping point: the decision to anticoagulate patients with atrialﬁbrillation. Circ Cardiovasc Qual Outcomes 2011;4:14-21.

[32] Liao J, Khalid Z, Scallan C, Morillo C, O'Donnell M. Noninvasive cardiac monitoring for detecting paroxysmal atrialﬁbrillation orﬂutter after acute ischemic stroke: a systematic review. Stroke 2007;38:2935-40.