1

Cover title: AF detection after TIA: systematic review/meta-analysis

Title: Cardiac monitoring for detection of atrial fibrillation after TIA: a systematic review and meta-analysis

Eleni Korompoki1 MD PhD, Angela Del Giudice1 MD, Steffi Hillmann2 MPH, Uwe Malzahn2,3 PhD, David J Gladstone4 MD, PhD, Peter Heuschmann2,3 MD, MPH, Roland Veltkamp1 MD

1 Division of Brain Sciences, Department of Stroke Medicine, Imperial College, London, UK

2 Institute of Clinical Epidemiology and Biometry, University of Würzburg, Germany

3 Clinical Trial Center , Comprehensive Heart Failure Center,UniversityHospital Würzburg, Germany

4 Division of Neurology, Department of Medicine, University Toronto, Ontario, Canada

Corresponding author:

Eleni Korompoki, MD PhD

Department of Stroke Medicine

Imperial College London

Fulham Palace Road

London, W6 8RF, UK

Phone +44-20-33130133

e.korompoki@imperial.ac.uk

Supplemental material (tables I-V, e-references, supplementary figure)

Tables: 2

Figures: 5

References: 55

Abstract: 245 words

Manuscript: 5141 words

Key words: ischemic attack, transient; atrial fibrillation; electrophysiology

2

Abstract

Background and Purpose: The detection rate of atrial fibrillation (AF) has not been studied

specifically in TIA patients although extrapolation from findings in ischemic stroke (IS) may

be inadequate. We conducted a systematic review and meta-analysis to determine the rate of

newly diagnosed AF using different methods of ECG monitoring in TIA.

Methods: A comprehensive literature search was performed following a pre-specified

protocol in accordance with the PRISMA statement. Prospective observational studies and

randomized controlled trials were considered that included TIA patients who underwent

cardiac monitoring for >12 hours. Primary outcome was frequency of detection of AF lasting

at least 30 sec. Analyses of subgroups and of duration and type of monitoring were

performed.

Results: 17 studies enrolling 1163 patients were included. Overall, the pooled AF detection

rate was 4% (95% CI: 2%-7%). Yield of monitoring was higher in selected (in terms of age,

pre-screening for arrhythmias and cause of TIA) than in unselected cohorts (7% vs 3%).

Pooled mean AF detection rates rose with duration of monitoring: 4% (24 hours), 5% (24

hours to7 days) and 6% (>7 days), respectively. The yield of non-invasive monitoring was

significantly lower than that of invasive monitoring (4% vs. 11%). Significant heterogeneity

was observed among studies (I2=60.61%).

Conclusion: This first meta-analysis of AF detection in TIA patients suggests that the rate of

AF detection is lower in TIA patients than reported for IS and TIA cohorts in previous meta-

analyses. Prospective studies are needed to determine the optimal diagnostic procedure for

AF detection in TIA.

3

Introduction

Sensitive diagnosing of atrial fibrillation (AF) represents a major objective in stroke and TIA

patients (1). While growing evidence guides diagnostic work-up for detection of AF in stroke

patients, the yield of various modalities of cardiac monitoring in TIA patients has not been

studied specifically (2, 3). Extrapolation of findings in stroke cohorts to TIA patients may be

inadequate. Although TIAs are considered as part of the spectrum of ischemic stroke (IS) (4),

the diagnosis of TIA is often based on medical history alone and is less definite. Moreover,

the setting of cardiac diagnostic work-up for strokes and TIAs differs substantially. In many

health care systems, TIAs are evaluated as outpatients (5-7) whereas strokes are admitted to

monitored stroke units (8, 9), Even if TIA patients are hospitalized, length of monitoring is

shorter (10). No systematic review assessing different techniques for AF detection in TIA

patients is available. Thus, the optimal duration, method and timing of monitoring and the

best candidates for prolonged monitoring in TIA are ill defined.

We performed a systematic review and meta-analysis to determine the overall rate of newly

detected AF after TIA. We also explored detection rates in selected (in terms of age, extent of

cardiac tests before enrollment, cryptogenic cause) and unselected TIA cohorts, and

evaluated the impact of duration and type of monitoring.

Methods

Search strategy and data extraction

We performed a systematic review following PRISMA guidance (figure 1)(11) and a pre-

specified study protocol. Relevant studies were identified using thesaurus subject headings

(MEDLINE and EMBASE) and free text terms (table s-1). Google scholar, Cochrane library,

clinical trial registries and references lists from all included articles and abstracts were also

4

assessed. Pre-defined eligibility criteria regarding study design, duration of monitoring, AF

definition were applied by two independent investigators (EK, SH). If necessary authors

were asked to provide information on eligibility criteria (e.g. study design, AF definition,

exclusion of known AF cases, duplicate publications). Corresponding authors of studies

including both IS and TIA patients were contacted by email to provide data about

characteristics and AF detection rates in TIA patients specifically. For studies which

examined more than one method of ECG monitoring in the same cohort, detection rates were

calculated for each method. Consequently, the number of pooled estimates exceeded the

number of eligible studies.

Eligibility criteria

We considered English and non-English articles and original abstracts published between

1966 and March 1, 2015. We included studies that enrolled TIA patients or mixed

populations of TIA and IS patients who underwent ECG monitoring for at least 12 hours.

Only randomized controlled trials and prospective observational studies reporting AF

episodes lasting ≥ 30 sec were included. Studies including patients with known AF or with

AF diagnosed during screening were excluded.

Outcome measures

The main outcome measure was the overall rate of newly detected AF. The proportion of

patients with AF in each subgroup was calculated by dividing the number of patients with

newly detected AF by the total number of patients included in this group.

Three subgroup analyses were performed. First, we determined the AF detection rate in

unselected versus selected patients. Included studies used various selection criteria. Some

included only “cryptogenic” TIAs or only supposedly embolic TIAs. Other selection criteria

5

related to duration and methods of screening for cardiac arrhythmias before enrolment. Some

studies only enrolled patients above a defined age. Second, we reported AF detection rates

depending on duration of ECG monitoring which was categorized into intervals ≤ 24 hours,

between 24 hours and 7 days, and > 7 days. Finally, we estimated the AF detection rate

depending on the non-invasive versus invasive monitoring).

Risk of bias assessment

We assessed the quality of the studies according to the Cochrane handbook (12). We used a

funnel plot and Egger regression test in order to assess publication bias (13). For the funnel

plot we plotted the AF detection rate (based on the double arcsine transformation) against the

precision (1/standard error). A p value <0.1 was considered significant for publication bias.

Statistical analysis

Four meta-analyses were performed to determine the overall AF detection rate in the total

population and in predefined subgroups. We used the random-effects model for all analyses

because a high degree of heterogeneity was expected among the included cohorts.

Confidence intervals (CI) were calculated by the exact binomial method. Freeman-Tukey

double arcsine transformation was used to stabilize the variance. Heterogeneity was assessed

by the χ2-based Q statistics, a measure of variance on a relative scale, and by Ι2, a quantitative

measure of inconsistency, which is neither sensitive to the metric of the analyzed quantity nor

to the number of studies.Within a random effects model, heterogeneity between study

subgroups was assessed using Q-statistics (expressed as “Heterogeneity between groups” in

forest plots). Higher values of I2 or values of p<0.1 for the Q-statistic based χ2-test suggest

significant heterogeneity. Pooled proportion meta-analysis was performed using STATA

(version 13.0) (14).

6

Results

Twenty four articles, 16 full manuscripts and 8 abstracts, of 1205 identified citations fulfilled

the eligibility criteria (figure 1) (3, 15-37). Three studies (13%) were randomized controlled

trials and 21 (87%) were prospective observational (table s-2). After 7 studies were excluded

from the analysis because of missing data, 17 studies were included in the meta-analysis.

Therein, 25 monitored study subgroups were reported because some studies investigated

various diagnostic procedures in the same cohort (figure 2). Baseline characteristics were

available for 14 studies (table s-2). Only one study was performed exclusively in TIA patients

(32). We excluded 147 patients because in 17 cases AF was detected at presentation (27, 37),

and for 130 patients the detection rate was not available (15, 18, 25, 30, 34-36). Thus, a total

of 1163 patients from 17 studies were eligible for analysis (table s-2). Since 7 of these studies

investigated more than one modality in the same study cohort, the number of included study

subgroups exceeded the number of eligible studies. Thus, 1423 data sets from 25 cohorts

were included in our meta-analysis (3, 16, 17, 19-24, 26-29, 31-33, 37).

Overall AF detection rate

The overall pooled proportion of TIA patients with newly detected AF was 4% (95% CI: 2%-

7%) (figure 2). The study reporting the highest yield for AF detection (27%, 95% CI: 8%-

55%) used inpatient ECG monitoring (29). No AF episodes were detected in seven other

cohorts using different modalities (19, 20, 22, 26, 29, 31, 33). Significant heterogeneity was

observed among studies (Ι2=60.61%, p=0.00) (figure 2).

Subgroup analyses

Selected versus unselected cohorts

7

Data regarding unselected and selected cohorts are presented in tables 1 and 2, respectively.

Selection criteria differed substantially among groups (table 2, table s-4). AF was more likely

to be uncovered in selected (7%, 95% CI: 2%-14%) compared to unselected patients (3%,

95% CI: 2%-5%) with significant heterogeneity between groups (p=0.041) (figure 3).

Duration of monitoring

Extending the duration of monitoring from ≤24 hours to 7 days conferred a 1% higher

detection rate (4%; CI: 1%-8% vs. 5%; 95% CI: 2%-9%). Further extension of monitoring

beyond 7 days resulted in another 1% increase of AF detection (6%, 95% CI: 1%-12%). No

significant heterogeneity was observed among groups (p=0.558) (figure 4).

Invasive versus Non-invasive monitoring

Invasive methods yielded a higher detection rate than non-invasive modalities (11%: 95% CI:

2%-24% vs. 4%, 95% CI: 2%-7%). Significant heterogeneity existed between groups

(p=0.05). The heterogeneity among studies using invasive methods of monitoring was 0%

(figure 5) but the small number of studies precluded an estimation of between-study variance

with acceptable precision. Moreover, I2 is truncated to zero when the Q statistic is smaller

than its degrees of freedom (38).

Assessment of publication bias

The funnel plot showed asymmetry, suggesting a possible publication bias. We acknowledge

that the interpretation of funnel plot asymmetry may be inaccurate for assessment of

publication bias in meta-analyses of proportions, particularly in case of low proportions (39).

In this setting, the choice for both vertical and horizontal axes is challenging, and it can lead

to misinterpretation of funnel plot asymmetry (40). To quantitatively assess the risk of bias,

we performed an Egger regression test which was not significant (p=0.552), thus not allowing

8

to reject the null hypothesis that there is no small study effect (figure s-1). Overall 82% of the

included studies had possible selection bias. Potential funding bias was found in 29% of the

studies. We anticipate lower rate of reporting bias as in the vast majority (70% of included

studies) AF detection rates were provided directly by the corresponding authors after personal

communication. Other risk of bias are presented in table s-5.

Discussion

The major findings of this study are: (1) The overall rate of newly detected AF in TIA

patients is 4%. (2) The AF detection rate was 7% in selected compared to 3% in unselected

cohorts. (3) Invasive monitoring yielded substantially higher rates of AF detection than non-

invasive methods. (4) The yield of monitoring for > 7d confers limited added value compared

to 24 hours. (5) Inpatient ECG monitoring resulted in a relatively high rate of AF detection.

The overall new AF detection rate in TIA patients of 4% in our review is substantially lower

than rates reported in previous reviews of studies including both IS and TIA patients (i.e. 6.3-

11.5%) (41-43). Previous meta-analyses revealed considerably higher heterogeneity assessed

by I2 (88.2to 90.12%) compared to our study (I2=60.61%) (41, 43). A potential explanation

for both lower AF detection rates and lower heterogeneity in our meta-analysis could be that

we used stricter inclusion criteria. TIA patients only represent a relatively small proportion of

patients enrolled in studies included in previous meta-analyses on both IS and TIA patients.

Differences of baseline characteristics between IS and TIA patients are another potential

cause of different detection rates. Moreover, the variability in TIA definitions across the

included studies may have led to an overestimation of the total number of TIA patients (table

s-3). For example, brain imaging was not required to rule out conditions mimicking TIAs in

some studies, and TIA mimics are not at the same risk for subsequent stroke as patients with

a confirmed diagnosis of TIA (44). We considered only studies with AF episodes lasting

9

more than 30 sec. In contrast, previous reviews including mixed IS and TIA cohorts

attributed part of the observed heterogeneity and bias to less strict AF definitions (41, 42). An

alternative explanation for the lower prevalence of AF in TIA patients is that a proportion of

newly detected AF in IS is attributed to the infarct (i.e. “neurogenic AF”). As the majority of

TIA patients have no or only small infarcts (45) AF is unlikely to result from TIA. Among

studies included into our review, only the CRYSTAL-AF trial included exclusively patients

with infarcts on brain imaging (46) whereas the remaining studies recruited patients based on

symptom-duration alone (4) (table s-4).

The average incidence of AF was more than twice as high in selected compared to unselected

TIA patients. The wide range of detection rates (0%-27%) and the higher heterogeneity in

selected cohorts may arise from selection criteria applied in this group. Patients were

recruited into specific studies based on age, availability of cardiac tests before screening for

AF, and presumed or “cryptogenic” etiology of TIA. Even in the “cryptogenic TIA” group

inclusion criteria varied from just negative ECG and 24-hour Holter to pre-specified

extensive diagnostic work-up before enrollment (table s-4). Moreover, the time interval

between TIA and initiation of monitoring during which the investigations were performed

differed substantially among studies. In eight studies TIA patients received prolonged

monitoring within the first 72 hours after the index event (3, 20, 21, 24, 27-29, 37) whereas

other studies recruited patients later than 30 days after TIA (16, 17, 19, 22, 23, 31). In some

studies monitoring started immediately after admission in an inpatient setting (3, 20, 21, 24,

27-29, 33, 36, 37) whereas patients received a monitor as outpatients in other studies (16-19,

22, 23, 25, 26, 31, 32, 34, 35). Nevertheless, AF detection is more likely in selected patients

with suspected embolism compared to unselected populations (23, 31, 47). Older age, certain

clinical and neuroimaging findings (e.g. cortical deficits and embolic lesion pattern,

respectively) (48-50), and cardiovascular characteristics (e.g. higher risk stratification scores,

10

atrial premature beats, enlarged left atrium) (51, 52) may identify patients at higher risk of

AF. Stratification according to risk may increase the yield of extensive diagnostic

investigation.

Our finding that invasive monitoring yields much higher detection rates than non-invasive

techniques (figure 4) should be interpreted with caution. First, the number of TIA patients

who received an insertable device was small (44 patients), and it is likely that TIA patients

receiving invasive monitoring were selected (table s-3). Two of the three studies assessing

invasive monitoring reported an AF detection rate of 15% (17, 31) whereas the third study

failed to uncover any AF but only 3 young TIA patients were included (19). Importantly, our

comparison between invasive and non-invasive methods of monitoring is limited because the

non-invasive group comprised a wide variety of monitoring periods (i.e. 24 hours to 30 days).

Because the importance of the duration of monitoring for AF detection has been

unequivocally shown in stroke patients, we performed a separate meta-analysis addressing of

the impact of different monitoring periods. This analysis yielded only a limit additional yield

of monitoring beyond 7 days. It remains to be shown whether very extensive non-invasive

recording as performed over 30 days in the EMBRACE trial in stroke and TIA patients

results in a substantially higher rate of AF detection in unselected TIA patients (23).

In unselected and selected TIA cohorts inpatient ECG monitoring yielded the highest rate of

AF detection (3, 29). Rapid cardiac assessment of TIA patients is more likely to occur in TIA

inpatients with a recent episode. Recently revised guidelines recommend prolonged

monitoring for AF after stroke or TIA if no cause is apparent (7, 47). However, the optimal

setting, method, duration and timing of monitoring in TIA patients remain unknown as

prospective studies specifically addressing TIA patients are lacking.

11

Our study has strengths and limitations. We reported an overall AF detection rate of 4%

however our results should be interpreted with caution because of significant heterogeneity

between studies: We combined studies with different designs, methods, modalities, duration

of monitoring and cohort characteristics. However we performed three different subgroup

analyses in order to evaluate the detection rates in selected and unselected cohorts as well as

to explore the influence of the duration and modality (invasive versus non invasive) on AF

detection rates after TIA. We performed a comprehensive search of the literature and

followed a strict search protocol consistent with PRISMA guidance. On the other hand, we

obtained data from only 17 of 24 eligible studies which may have led to a selection bias.

However, the included studies reported on 89% of patients enrolled in all studies. Studies

which examined more than one modality for detecting AF in the same patient population (3,

20, 29) were considered as different independent studies in our meta-analysis as in previous

meta-analyses on the topic (41, 43). This may have resulted in a spuriously small estimate of

the variance of the effect size and in a high likelihood of a type I error. As sensitivity

analyses including only one modality for each study at a time provided similar results (data

not shown), we did not apply more complex methods of dealing with dependency such as

multi-level meta-analysis and robust variance estimation which also have limitations (53, 54).

Furthermore, we were unable to assess the interval between the index event and the initiation

of monitoring. However, our meta-analysis suggests that inpatient monitoring has a higher

yield of AF detection in both selected and unselected patients. Funnel plot showed

asymmetry suggesting a risk of bias. However, we tried to minimize publication bias by

searching all potential studies independently of the language and type of publication in

different databases. Finally, because most studies did not provide baseline characteristics for

risk stratification including age, gender and ABCD2 score further subgroup analyses were not

possible.

12

Summary/Conclusions

This systematic review and meta-analysis finds a substantially lower detection rate of AF in

TIA patients compared to previous analyses of mixed cohorts of predominantly stroke and

less TIA patients. Additional analyses suggest that selected patients or those undergoing

invasive monitoring have higher rates of AF detection than unselected patients. As

extrapolation from findings in stroke cohorts to TIA patients may be inadequate, prospective

studies are needed to determine the most appropriate method and duration of prolonged

cardiac monitoring in TIA patients specifically.

Acknowledgements

We are grateful to Dr Dominique Babuty, Dr Luisa Christensen, Dr Exuperio Díez Tejedor,

Dr Catarina Fonseca, Dr Hooman Kamel, Dr Patricia Martinez Sanchez, Dr Gerben Plas, Dr

Timolaos Rizos, Dr Wadea Tarhuni, Dr Vincent Thijs, Dr Rolf Wachter, Dr Marianne Zeller,

who provided data from their studies for this analysis.

We would also like to thank Dr Manuel Jesus Gomez-Choco for his help with translations.

None of these individuals had access to the manuscript before submission and they are not

responsible for its contents.

Sources of Funding

St. Mary´s development fund, National Institute for Health Research (NIHR) Imperial

Biomedical Research Centre.

13

Conflicts of Interest/|Disclosures

EK, ADG, SH, UM: None

DG: speaker honoraria, advisory boards: Bayer, Boehringer Ingelheim, BMS, Daiichi-

Sankyo, Pfizer.

PH: research grants BMBF, European Union, Charité, Berlin Chamber of Physicians,

German Parkinson Society, UniversityHospital Würzburg, Robert-Koch-Institute, Charité–

Universitätsmedizin Berlin (MonDAFIS,unrestricted research grant to Charité, Bayer),

University Göttingen (FIND-AFrandomized, unrestricted research grant, University

Göttingen, Boehringer-Ingelheim), UniversityHospital Heidelberg (RASUNOA-prime;

unrestricted research grant , Bayer, BMS, Boehringer, Daiichi Sankyo), outside submitted

work.

RV: Consulting, speaker honoraria, research support: Bayer, Boehringer , Daiichi Sankyo,

BMS, Pfizer, Biogen, Morphosys, Medtronic, Apoplex Medical technologies

14

References

1. 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(12):857-67.2. Kallmunzer B, Breuer L, Hering C, et al. A structured reading algorithm improves telemetric detection of atrial fibrillation after acute ischemic stroke. Stroke. 2012;43(4):994-9.3. Rizos T, Güntner J, Jenetzky E, et al. Continuous stroke unit electrocardiographic monitoring versus 24-hour Holter electrocardiography for detection of paroxysmal atrial fibrillation after stroke. Stroke. 2012;43(10):2689-94.4. Easton JD, Saver JL, Albers GW, et al. Definition and evaluation of transient ischemic attack: a scientific statement for healthcare professionals from the American Heart Association/American Stroke Association Stroke Council; Council on Cardiovascular Surgery and Anesthesia; Council on Cardiovascular Radiology and Intervention; Council on Cardiovascular Nursing; and the Interdisciplinary Council on Peripheral Vascular Disease. The American Academy of Neurology affirms the value of this statement as an educational tool for neurologists. Stroke. 2009;40(6):2276-93.5. Rothwell PM, Giles MF, Chandratheva A, et al. Effect of urgent treatment of transient ischaemic attack and minor stroke on early recurrent stroke (EXPRESS study): a prospective population-based sequential comparison. Lancet. 2007;370(9596):1432-42.6. Lavallee PC, Meseguer E, Abboud H, et al. A transient ischaemic attack clinic with round-the-clock access (SOS-TIA): feasibility and effects. Lancet Neurol. 2007;6(11):953-60.7. Verma A, Cairns JA, Mitchell LB, et al. 2014 focused update of the Canadian Cardiovascular Society Guidelines for the management of atrial fibrillation. Can J Cardiol. 2014;30(10):1114-30.8. Jauch EC, Saver JL, Adams HP, Jr., et al. Guidelines for the early management of patients with acute ischemic stroke: a guideline for healthcare professionals from the American Heart Association/American Stroke Association. Stroke. 2013;44(3):870-947.9. Guidelines for management of ischaemic stroke and transient ischaemic attack 2008. Cerebrovasc Dis. 2008;25(5):457-507.10. 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(6):1693-5.11. Liberati A, Altman DG, Tetzlaff J, et al. The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: explanation and elaboration. J Clin Epidemiol. 2009;62(10):e1-34.12. Higgins JPT AD. Assessing risk of bias in included studies : in Higgins JPT GS, editor: Chichester, UK, Wiley; 2008. 1-53 p.13. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. Bmj. 1997;315(7109):629-34.14. Nyaga VN, Arbyn M, Aerts M. Metaprop: a Stata command to perform meta-analysis of binomial data. Arch Public Health. 2014;72(1):39.15. Barthelemy JC, Feasson-Gerard S, Garnier P, et al. Automatic cardiac event recorders reveal paroxysmal atrial fibrillation after unexplained strokes or transient ischemic attacks. Ann Noninvasive Electrocardiol. 2003;8(3):194-9.16. Callero EC, Martinez-Sanchez P, Gomar DP, et al. Utility a second 24-hours holter monitoring for the diagnosis of paroxismal atrial fibrillation after an acute stroke. Stroke. 2012;43 (2 Meeting Abstracts).17. Christensen LM, Krieger DW, Hojberg S, et al. Paroxysmal atrial fibrillation occurs often in cryptogenic ischaemic stroke. Final results from the SURPRISE study. Eur J Neurol. 2014;21(6):884-9.

15

18. Deelawar S, Manegold JC, Kalyani M, et al. Diagnostic yield of extended-time external loop recording for the detection of paroxysmal atrial fibrillation in patients with cryptogenic stroke. Europace. 2013;15:ii220.19. Dion F, Saudeau D, Bonnaud I, et al. Unexpected low prevalence of atrial fibrillation in cryptogenic ischemic stroke: a prospective study. J Interv Card Electrophysiol. 2010;28(2):101-7.20. Dobbels L, Thijs V. Prospective comparison of continuous cardiac monitoring and Holter on a stroke unit. Eur J Neurol. 2014;21:116.21. Fernandez V, Bejot Y, Zeller M, et al. Silent atrial fibrillation after ischemic stroke or transient ischemic attack: interest of continuous ECG monitoring. Eur Neurol. 2014;71(5-6):313-8.22. Fonseca AC, Brito D, Pinho e Melo T, et al. N-terminal pro-brain natriuretic peptide shows diagnostic accuracy for detecting atrial fibrillation in cryptogenic stroke patients. Int J Stroke. 2014;9(4):419-25.23. Gladstone DJ, Spring M, Dorian P, et al. Atrial fibrillation in patients with cryptogenic stroke. N Engl J Med. 2014;370(26):2467-77.24. 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(12):3357-64.25. Jabaudon D, Sztajzel J, Sievert K, Landis T, Sztajzel R. Usefulness of ambulatory 7-day ECG monitoring for the detection of atrial fibrillation and flutter after acute stroke and transient ischemic attack. Stroke. 2004;35(7):1647-51.26. Kamel H, Navi BB, Elijovich L, et al. Pilot randomized trial of outpatient cardiac monitoring after cryptogenic stroke. Stroke. 2013;44(2):528-30.27. Martinez-Sanchez P, Callero EC, Herranz AC, et al. Detection of paroxysmal atrial fibrillation in patient with acute brain ischemia combining cardiac and holter monitoring: Prevalence and predictors. Stroke. 2012;43 (2 Meeting Abstracts).28. Plas GJ, Bos J, Velthuis BO, Scholten MF, den Hertog HM, Brouwers PJ. Diagnostic yield of external loop recording in patients with acute ischemic stroke or TIA. J Neurol. 2015;262(3):682-8.29. Rizos T, Rasch C, Jenetzky E, et al. Detection of paroxysmal atrial fibrillation in acute stroke patients. Cerebrovasc Dis. 2010;30(4):410-7.30. Rodriguez-Leal C, Toro-Cebada R, Quezada Feijoo M, Cabeza-Letran L. Holter 24-hours ECG and clinical predictors for arrhythmia detection in patients with cardioembolic stroke profile. Thromb Res. 2014;133:S115.31. Sanna T, Diener HC, Passman RS, et al. Cryptogenic stroke and underlying atrial fibrillation. N Engl J Med. 2014;370(26):2478-86.32. Tarhuni W, El-Masri M. A Double-blind randomized controlled trial comparing the effectiveness of external loop monitor and automatic trigger atrial fibrillation memory loop recorder in detecting atrial fibrillation among patients with tia: A pilot study. Crit Pathw Cardiol. 2014;13 (3):130.33. Thakkar S, Bagarhatta R. Detection of paroxysmal atrial fibrillation or flutter in patients with acute ischemic stroke or transient ischemic attack by Holter monitoring. Indian Heart J. 2014;66(2):188-92.34. Ungar A, Rieger G, West T, et al. Atrial fibrillation and treatment changes in cryptogenic stroke patients with an implantable loop recorder for continuous cardiac rhythm monitoring. Eur Heart J. 2013;34:62.35. Ustrell X, Pellise A, Merce J, Vinas J, Mares R. Feasibility of implantable cardiac monitors for the detection of occult atrial fibrillation in embolic cryptogenic stroke. Cerebrovascular Dis. 2012;33:376-7.36. Vivanco Hidalgo RM, Rodriguez Campello A, Ois Santiago A, Cuadrado Godia E, Pont Sunyer C, Roquer J. Cardiac monitoring in stroke units: importance of diagnosing atrial fibrillation in acute ischemic stroke. Rev Esp Cardiol. 2009;62(5):564-7.

16

37. Wachter R, Weber-Kruger M, Seegers J, et al. Age-dependent yield of screening for undetected atrial fibrillation in stroke patients: the Find-AF study. J Neurol. 2013;260(8):2042-5.38. Huedo-Medina TB, Sanchez-Meca J, Marin-Martinez F, Botella J. Assessing heterogeneity in meta-analysis: Q statistic or I2 index? Psychol Methods. 2006;11(2):193-206.39. Hunter JP, Saratzis A, Sutton AJ, Boucher RH, Sayers RD, Bown MJ. In meta-analyses of proportion studies, funnel plots were found to be an inaccurate method of assessing publication bias. J Clin Epidemiol. 2014;67(8):897-903.40. Sterne JA, Egger M. Funnel plots for detecting bias in meta-analysis: guidelines on choice of axis. J Clin Epidemiol. 2001;54(10):1046-55.41. 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(2):520-6.42. Seet RC, Friedman PA, Rabinstein AA. Prolonged rhythm monitoring for the detection of occult paroxysmal atrial fibrillation in ischemic stroke of unknown cause. Circulation. 2011;124(4):477-86.43. Dussault C, Toeg H, Nathan M, Wang ZJ, Roux JF, Secemsky E. Electrocardiographic monitoring for detecting atrial fibrillation after ischemic stroke or transient ischemic attack: systematic review and meta-analysis. Circ Arrhythm Electrophysiol. 2015;8(2):263-9.44. Dutta D, Bowen E, Foy C. Four-year follow-up of transient ischemic attacks, strokes, and mimics: a retrospective transient ischemic attack clinic cohort study. Stroke. 2015;46(5):1227-32.45. Brazzelli M, Chappell FM, Miranda H, et al. Diffusion-weighted imaging and diagnosis of transient ischemic attack. Ann Neurol. 2014;75(1):67-76.46. Sinha AM, Diener HC, Morillo CA, et al. Cryptogenic Stroke and underlying Atrial Fibrillation (CRYSTAL AF): design and rationale. American Heart Journal.160(1):36-41.e1.47. Kernan WN, Ovbiagele B, Black HR, et al. Guidelines for the prevention of stroke in patients with stroke and transient ischemic attack: a guideline for healthcare professionals from the American Heart Association/American Stroke Association. Stroke. 2014;45(7):2160-236.48. Kimura K, Minematsu K, Wada K, Yonemura K, Nakajima M. Clinical characteristics in transient ischemic attack patients with atrial fibrillation. Intern Med. 2003;42(3):255-8.49. Purroy F, Montaner J, Molina CA, Delgado P, Ribo M, Alvarez-Sabin J. Patterns and predictors of early risk of recurrence after transient ischemic attack with respect to etiologic subtypes. Stroke. 2007;38(12):3225-9.50. Redgrave JN, Coutts SB, Schulz UG, Briley D, Rothwell PM. Systematic review of associations between the presence of acute ischemic lesions on diffusion-weighted imaging and clinical predictors of early stroke risk after transient ischemic attack. Stroke. 2007;38(5):1482-8.51. Gladstone DJ, Dorian P, Spring M, et al. Atrial premature beats predict atrial fibrillation in cryptogenic stroke: results from the EMBRACE trial. Stroke. 2015;46(4):936-41.52. Weber-Kruger M, Groschel K, Mende M, et al. Excessive supraventricular ectopic activity is indicative of paroxysmal atrial fibrillation in patients with cerebral ischemia. PLoS One. 2013;8(6):e67602.53. Scammacca N, Roberts G, Stuebing KK. Meta-Analysis With Complex Research Designs: Dealing With Dependence From Multiple Measures and Multiple Group Comparisons. Rev Educ Res. 2014;84(3):328-64.54. Van den Noortgate W, Lopez-Lopez JA, Marin-Martinez F, Sanchez-Meca J. Three-level meta-analysis of dependent effect sizes. Behav Res Methods. 2013;45(2):576-94.

17

Tables

Table 1. AF detection rates in unselected patients.

*,**same population

CEM: continuous ECG monitoring (inpatient), CEMa: automatically analysed continuous ECG monitoring, MCOT: mobile cardiac output telemetry ELR: external loop recorder,

Study N eligible for analysis

Detection method

Duration of monitoring

Detection rate of new AF, (95% CI)

Dobbels, 2014a* 16 Holter 24h 6.3%

Dobbels, 2014b* 16 CEM 24h 0.0%

Fernandez, 2014 35 CEM ≥24h 5.7%

Grond, 2013 326 Holter 72h 2.8%

Martinez-Sanchez, 2012 114 CEM 48h-72h 0.8%

Rizos, 2012a** 97 Holter 24h 6.2%

Rizos, 2012b** 97 CEM >24h 7.2%

Rizos, 2012c** 97 CEMa >24h 8.2%

Tarhuni, 2014a 80 ELR 14d 3.8%

Tarhuni, 2014b 80 MCOT 14d 3.8%

Thakkar, 2014 10 Holter 24h 0.0%

Wachter, 2013 72 Holter 7d 4.2%

18

Table 2. AF detection rates in selected patients.

Study N Selection criteria MethodOf monitoring

Duration of monitoring

New AF,%

Callero, 2012 19 ECG and 1st 24h Holter: negative

Holter 24h 15.8%

Christensen, 2013 20 Cryptogenic ILR 2y (702d) 15.0%

Dion, 2010 3 Cryptogenic, age<75y, controlled HTN

ILR 14m 0.0%

Fonseca, 2014 5 Cryptogenic Holter 24h (3m and 6m)

0.0%

Gladstone, 2014a 110 Cryptogenic, age>55y

Holter 24h 0.9%

Gladstone, 2014b 98 Cryptogenic, age>55y

ELR 30d 15.3%

Kamel, 2013 8 Cryptogenic, high risk TIA (ABCD2≥4)

MCOT 21d 0.0%

Martinez-Sanchez, 2012b

37 TIA of unknown origin or suspected cardiac embolism

Holter 24h 18.9%

Plas, 2015 15 ECG and 24h telemetry: negative

ELR 24h 13.3%

Rizos, 2010a* 13 Age>60y Holter 24h 0.0%

Rizos, 2010b* 15 Age>60y CEM 48h 26.7%

Sanna, 2014a 21 Cryptogenic, age>40y

ICM 6months 15.0%

Sanna, 2014b 19 Cryptogenic, age>40y

Holter 24 hours 0.0%

*same population

ECG: electrocardiogram, ILR: implantable loop recorder, ELR: external loop recorder, CEM: continuous ECG monitoring, ICM: implantable cardiac monitor, HTN: hypertension

19

Figures

Figure 1. PRISMA flow chart

20

Figure 2. Forrest plot showing proportion of TIA patients diagnosed with AF in all included cohorts

using different modalities of ECG monitoring. ES: effect size, CI: confidence intervals

21

Figure 3. Forrest plot showing proportion of TIA patients diagnosed with AF in selected and

unselected cohorts. ES: effect size, CI: confidence intervals

22

Figure 4. Forrest plot showing proportion of TIA patients diagnosed with AF depending on duration

of ECG monitoring. Duration of monitoring was categorized into <=24 hours, >24 hours to 7 days, >

7 days). ES: effect size, CI: confidence intervals

23

Figure 5. Forrest plot with proportion of TIA patients diagnosed with AF using non-invasive versus

invasive methods of ECG monitoring. ES: effect size, CI: confidence intervals