Excessive Supraventricular Ectopic Activity Is Indicative of Paroxysmal Atrial Fibrillation in Patients with Cerebral Ischemia

Background Detecting paroxysmal atrial fibrillation (PAF) in patients with cerebral ischemia is challenging. Frequent premature atrial complexes (PAC/h) and the longest supraventricular run on 24-h-Holter (SV-run24 h), summarised as excessive supraventricular ectopic activity (ESVEA), may help selecting patients for extended ECG-monitoring, especially in combination with echocardiographic marker LAVI/a’ (left atrial volume index/late diastolic tissue Doppler velocity). Methods Retrospective analysis from the prospective monocentric observational trial Find-AF (ISRCTN-46104198). Patients with acute stroke or TIA were enrolled at the University Hospital Göttingen, Germany. Those with sinus rhythm at presentation received 7-day Holter-monitoring. ESVEA was quantified in one 24-hour interval free from PAF. Echocardiographic parameters were assessed prospectively. Results PAF was detected in 23/208 patients (11.1%). The median was 4 [IQR 1; 22] for PAC/h and 5 [IQR 0; 9] for SV-run24 h. PAF was more prevalent in patients with ESVEA: 19.6% vs. 2.8% for PAC/h >4 vs. ≤4 (p<0.001); 17.0% vs. 4.9% for SV-run24 h >5 vs. ≤5 beats (p = 0.003). Patients with PAF showed more supraventricular ectopic activity: 29 PAC/h [IQR 9; 143] vs. 4 PAC/h [1]; [14] and longest SV-run24 h = 10 [5]; [21] vs. 0 [0; 8] beats (both p<0.001). Both markers discriminated between the PAF- and the Non-PAF-group (area under receiver-operator-characteristics-curve 0.763 [95% CI 0.667; 0.858] and 0.716 [0.600; 0.832]). In multivariate analyses log(PAC/h) and log(SV-run24 h) were independently indicative of PAF. In Patients with PAC/h ≤4 and normal LAVI/a’ PAF was excluded, whereas those with PAC/h >4 and abnormal LAVI/a’ showed high PAF-rates. Conclusions ESVEA discriminated PAF from non-PAF beyond clinical factors including LAVI/a’ in patients with cerebral ischemia. Normal LAVI/a’+PAC/h ≤4 ruled out PAF, while prevalence was high in those with abnormal LAVI/a’+PAC/h >4.


Introduction
Though one of the most common causes of ischemic stroke, atrial fibrillation in its paroxysmal form can be challenging to diagnose. Bursts of paroxysmal atrial fibrillation (PAF) can be short, they often occur in clusters [1] and are asymptomatic in over 50% of the cases [2]. Diagnostic yield has been shown to increase with duration of monitoring [3][4][5]. Methods for prolonging ECG-monitoring over wider lengths of time have been developed [6]. Nevertheless, extended diagnostics are timeconsuming, expensive and cumbersome for both patients and investigators. Therefore it is essential to identify indicators of underlying PAF in patients with cerebral ischemia to focus extended diagnostics on high-risk patients.
As one potential approach, we have demonstrated that morphologic indicators of structural left atrial remodelling in echocardiography can be useful to rule out the presence of PAF within this collective [7]. As a different approach, biomarkers (e. g. natriuretic peptides) may serve to identify patients with PAF [8].
In addition to structural alterations, atrial fibrillation (AF) is also known to be accompanied by a state of electrical instability of the atrium [9]. There are some studies indicating that premature atrial complexes (PAC) [10][11][12][13] as well as supraventricular (SV) runs [14][15] are associated with prevalent or incident AF.
In a retrospective analysis of data from a prospective observational trial including patients with cerebral ischemia, we evaluated the association of frequent PAC and SV-runs on 24hour Holter as markers of atrial electric instability with the finding of PAF within a prolonged monitoring spell (7-day Holtermonitoring). We also evaluated previously defined cut-offs of comparable trials by Wallmann et al. [13] and Binici et al. [15]. Once the association was confirmed, we combined these electrocardiographic markers with a recently evaluated sensitive echocardiographic indicator of paroxysmal atrial fibrillation [7], equitable with structural myocardial remodelling (left atrial volume index/late diastolic mitral annular velocity = LAVI/a') to detect those patients after cerebral ischemia with a high probability of underlying PAF.

Study Design and Ethical Statement
The prospective monocentric observational trial Find-AF (ISRCTN 46104198) complies with the declaration of Helsinki, was performed according to the principles of ICH-GCP and was approved of by the local Ethics Committee at the University of Göttingen. Patients were only included after written and informed consent had been obtained.

Patients
Patients .18 years with suspected cerebral ischemia presenting at the emergency ward of the Göttingen University Hospital were prospectively included. Details on the cohort, detection of PAF by 7-day Holter-monitoring [5], as well as echocardiographic data to predict the presence of PAF [7] have been previously published. The aim of the study was to detect various factors associated with atrial fibrillation in patients with cerebral ischemia to guide future diagnostic pathways as well as therapeutic approaches. Cardiovascular endpoints were prospectively collected at a clinical followup visit after one year.

ECG Monitoring and Analysis
All patients presenting in sinus rhythm received prolonged ECG-monitoring for up to 7 days. The devices were applied early, i.e. at the emergency ward. The Holter-ECG recordings were analysed using the dedicated analysis-software (CardioDayH, getemed Medizin-und Informationstechnik, Teltow, Germany). The workup of the recorded ECG-data took place in two steps: One 24-hour period was analysed in full detail and was forwarded to the treating physician in a timely manner to represent the standard of care. We arbitrarily chose day four and deviated from this day only if the specific recording was unavailable or of explicitly bad quality. The remaining ECG-material was analysed at a later point in time with a clear focus on detecting PAF. The patient's general practitioner was contacted only in cases when AF (or another obvious heart rhythm disorder) was detected within the prolonged ECG-spell.
PAF was defined as an absolute arrhythmia .30 s without detectable P-waves and without a more likely differential diagnosis.
We calculated the number of PAC per hour and identified the longest SV-run within 24 hours (SVrun 24 h ) as a semi-quantitative surrogate for electrical instability (or excessive supraventricular ectopic activity = ESVEA) of the atrium. In most cases, we used the recording that had been thoroughly evaluated as part of the clinical routine for these analyses. If this specific recording contained an episode of PAF, we reanalysed the next day free from the arrhythmia.
To determine the number of PAC, all automatically detected events were manually reviewed and reclassified if necessary. All supraventricular runs .5 beats were counted manually to determine the longest episode within the 24-hour interval.

Echocardiography
As previously reported in detail [7], all transthoracic echocardiograms within Find-AF were performed by study personnel and a comprehensive dataset was prospectively collected. An index specifying left atrial contractility (left atrial volume index as calculated by the ellipsoid formula divided by late diastolic mitral annular velocity as an indicator of atrial contraction = LAVI/a') was found to be the strongest echocardiographic marker to rule out the presence of PAF and was therefore used for the multivariate analyses in this study.

Statistics
The prevalence of PAF in patients with and without ESVEA was compared using the Chi-Square-test. Differences in PACfrequency and the number of beats of the longest SV-run 24 h in the PAF vs. the Non-PAF-group were tested using the Mann-Whitney-U-test. The discrimination of ESVEA-parameters between both groups was determined using receiver-operator-characteristic (ROC) curves. We looked for potential influence factors by logistic regression, using age, sex, body-mass-index, systolic blood pressure, hypertension, heart failure and LAVI/a' as covariates.
The clinical parameters used in our analysis had been identified as predictors of incident atrial fibrillation in the Framingham study [16]. PAC/h and SV-run 24 h were log-transformed before statistical testing due to skewed distributions. We further included LAVI/a', the strongest echocardiographic marker for PAF in our study, as we hypothesised that the predictive value of this morphologic indicator of structural atrial remodelling might be distinct from that of ESVEA, an indicator of electrical atrial remodelling. In the case of separation (monotone likelihood), we estimated the effects by Firth's penalised-likelihood logistic regression following Heinze and Schemper [17]. For this purpose we applied the R package logistf [18]. All other analyses were performed by SPSS Statistics version 20. In general, tests were performed as two-sided at 5% significance level.

Results
229 patients presented in sinus rhythm and therefore underwent 7-day Holter-monitoring within the Find-AF-trial [5]. ESVEAanalysis could not be accomplished in 21 cases (12 pacemaker rhythms, 5 with insufficient quality, 4 without any PAF-free 24hour interval). The selection of patients for ESVEA-analysis is also shown in the study flow diagram (figure 1). Clinical characteristics are shown in table 1. LAVI/a' was available in 169 patients of the analysis population.
The predictive value of the ESVEA parameters was confirmed by repeating the analyses, taking into account only cases whose ESVEA parameters had been determined in an ECG-interval before first documentation of PAF. The revised median frequency of PAC remained 4/h and PAF prevalence remained significantly higher in those with PAC/h .4 (11.8%; CI 5.4%-18.7% vs. 1.9%; CI 0.0%-4.8%, p = 0.005). The revised median for longest SV-run 24 h , however, turned out to be 0 with the smaller number of PAF-patients and we did not apply this cut-off to calculate the differences in PAF prevalence. ESVEA rates remained higher in the PAF vs. the non-PAF group: the median was 11 (IQR 5/106) vs. 4 (IQR 1/15) for PAC/h and 12 (IQR 5/12) vs. 0 (IQR 0/8) for the longest SV-run 24 h . Both markers still discriminated between the PAF and the non-PAF group: The area under the ROC curve was 0.686 (CI 0.540-0.832), p = 0.038 for PAC/h and 0.733 (CI 0.575-0.891), p = 0.01 for the longest SV-run 24 h .
In multivariate analyses using age, sex, body-mass index, systolic blood pressure, hypertension and heart failure as covariates, log(PAC/h) (p = 0.002) and log(SV-run 24 h ) (p = 0.007) were independently associated with PAF (table S1A in file S1). The same was observed in a different model based on baseline differences between the groups (table S1B in file S1) as well as in a combination of both models (table S2 in file S1). Corrected for other significant predictors, patients with PAC/h .4 had an odds ratio of 5.7 (p = 0.011) and those with SV-run 24 h .5 beats of 3.1 (p = 0.054) of having PAF (table S3 in file S1). Importantly, both ESVEA parameters maintained significance when LAVI/a', the strongest echocardiographic predictor, was entered into the model (table S4A in file S1), with an adjusted odds ratio of 4.4 (p = 0.048) for PAC/h and 3.1 (p = 0.109) for SV-run 24 h above the median (table S4B in file S1).
Among those patients for whom LAVI/a' was available, 8 (4.7%) experienced a recurrent stroke during follow-up. Seven of those (87.5%) had initially shown a pathologic LAVI/a'-index, whereas only 3 (37.5%) had had frequent PAC.

Discussion
Our results imply that patients with ESVEA on 24-hour Holter-ECG show significantly higher PAF-rates within the entire, prolonged monitoring spell. Both parameters, frequent PAC and longest SV-run 24 h , are significantly and independently associated with underlying PAF. When combining PAC/h, a marker of atrial electric remodelling, with LAVI/a', an echocardiographic marker of structural atrial remodelling, PAF was literally ruled out in patients with normal results.

Association of ESVEA and PAF
Several investigators found that episodes of AF are often initiated by premature atrial complexes [19][20][21]. Some previous publications have examined the association of excessive supraventricular ectopic activity and the finding of atrial fibrillation. Healthy subjects with high rates of PAC or longer SV-runs on 24hour Holter were shown to be at a higher risk of developing atrial fibrillation or experiencing cerebral ischemia [14][15]. Retrospective analyses of Holter-ECGs recorded from patients after ischemic strokes showed a correlation of frequent PAC and finding of PAF [10][11]. Wallmann et al. analysed frequent PAC in patients with acute cerebral ischemia as an indicator of underlying PAF. Detection rates on delayed extended monitoring were significantly higher in those with more than approximately 3 PAC/h (70 PAC/ d) on an initial 24-hour Holter-ECG [12][13]. Our data are in good agreement with Wallmann's findings: In our study population, those with .4 PAC/h (median) in a 24-hour ECG-recording  [15], the applied thresholds appear to be too strict for patients with cerebral ischemia, as sensitivity is greatly reduced, while gaining only moderately increased specificity. Altogether, PAC/h appeared to show a more significant association with PAF and the combination with the longest SV-run 24 h did not improve the results.

Electrocardiographic and Echocardiographic Markers Help to Identify Patients at High Risk of Underlying PAF
We expected electrocardiographic/echocardiographic markers, both indicators of atrial pathologies, to be considerably more specific surrogates of atrial fibrillation compared to unspecific clinical parameters currently taken into account [22][23].
Multivariate analysis containing both PAC/h and LAVI/a' confirmed that both parameters are independently associated with PAF, possibly because structural and electrical remodelling occur separately, depending on the patient's individual AF-aetiology. Interestingly, PAF was literally ruled out in those who showed neither high rates of PAC on 24-hour Holter-ECG, nor signs of structural remodelling in echocardiography. Therefore it makes sense to combine both indicators to rule out atrial fibrillation in those without structural or electrical atrial alterations.

ESVEA and LAVI/a' as Prognostic Markers?
Within the Find-AF-trial we also reviewed cardiovascular events in up to one year of follow-up. Interestingly, our data suggest LAVI/a' may be a risk factor for recurrent strokes, while ESVEA, albeit similarly a surrogate for PAF in our cohort, does not appear to hold predictive value here. This might indicate that left atrial mechanical dysfunction per se is a risk factor for recurrent thromboembolism, while electrical instability is of lesser impor-tance. However, this analysis is limited by small numbers and can be considered as hypothesis-generating at best.

Clinical Relevance and Perspective
Compared to other common aetiologies of cerebral ischemia, paroxysmal AF is difficult to diagnose. Apart from being asymptomatic in a high percentage of cases [2], AF-bursts are often of short duration (while even short spells significantly increase stroke risk [24]) and furthermore they often appear in clusters (bearing the possibility of long periods free from the arrhythmia). Therefore PAF-detection-rates significantly increase by prolonging the monitoring spell.
Detection of PAF, however, is highly relevant, because oral anticoagulation has been shown to reduce recurrent ischemic events significantly, along with improving long-term prognosis [25][26]. Throughout the last years, methods for AF-detection have constantly improved [6]. New developments include implantable loop-recorders, that allow significantly longer monitoring periods and therefore offer the opportunity for presumably more sensitive AF-detection. While externally applied devices are generally cumbersome for patients and often bear a high workload for investigators, implanted devices require invasive procedures for implantation and explanation and all procedures are considerably expensive, limiting a wide spread utilisation in clinical practice.
This calls for valid and easily acquired markers of underlying PAF in patients with recent cerebral ischemia to increase the pretest probability in order to focus elaborate extended monitoring procedures on patients at a high risk of underlying AF. Several factors associated with AF have been identified [16] and risk scores have been published [22][23], the latter leaving room for further improvement [27]. Most importantly, we would like to point out, that many of the markers currently taken into account (e.g. age, gender, hypertension, cardiovascular disease, BMI) are very unspecific of PAF, rather indicating cardiovascular disease in general.
In search of more specific markers, we wanted to focus on indicators of atrial pathologies as supposedly better surrogates for atrial fibrillation. We used electrocardiographic parameters indicating electrical remodelling (frequent PAC and longer duration of short SV-runs summarised as ESVEA) as well as echocardiographic markers of structural remodelling of the atrium Figure 2. Duration of PAF episodes. Distribution of patients with longest episode of atrial fibrillation in 7-day Holter-monitoring in the respective category. Divisions were chosen based on trials indicating an increased thromboembolic risk above the respective cut-off: 6 min. as shown to increase stroke risk within the ASSERT-trial [24], 6 hours (5.5 hours) as presented by the TRENDS-study investigators [28] and 24 hours as identified within the Italian AT-500 registry [29]. 30 seconds as defined as minimal duration of atrial fibrillation by current AF guidelines [30]. doi:10.1371/journal.pone.0067602.g002 (LAVI/a'), both processes known to be associated with atrial fibrillation [9]. Importantly, both these markers are easily attained by standard clinical procedures performed after ischemic stroke and may be used complementary, depending on local availability.
Based on our findings, electrocardiographic and echocardiographic markers could help future development/improvement of predictive scores for underlying PAF in patients with acute cerebral ischemia. Their aim will be to select those with a high probability of PAF for extended ECG-monitoring.

Limitations
ESVEA was analysed retrospectively and were not a primary goal of the study. Furthermore, the 24-hour Holter-interval used for ESVEA-analysis was taken from the same, prolonged spell used for PAF detection. In our main analysis we investigated the association of ESVEA with PAF occurring both before and after ESVEA analysis as the best indicator of truly underlying PAF, which per se is not fully identical to predicting future detection (and possibly new incidence) of PAF. Although we confirmed the predictive value of ESVEA after excluding all cases of PAF that had occurred before ESVEA analysis, this confirmatory analysis is limited by the reduced number of PAF cases.

Summary
Electrocardiographic markers of atrial electric remodelling, PACs/h and longest SV-run 24 h , could be useful tools to identify patients with recent cerebral ischemia at high risk of underlying paroxysmal atrial fibrillation. In our cohort of patients with acute cerebral ischemia, PAF was ruled out in those without ESVEA on 24-hour Holter-ECG and with normal LAVI/a', an echocardiographic marker of structural remodelling.

Supporting Information
File S1 Multiple Binary Regression Models. Table S1A. Co-variables based on a predictive model for incident atrial fibrillation (Schnabel et al. [16]). All co-variables forced into the model. Table S1B. Co-variables based on significant baseline differences between groups. All co-variables forced into the model. Table S2. Co-variables based on significant contributors in models in tables S1A & S1B. Stepwise inclusion due to number of co-variables for model S2/1, forced inclusion for models S2/2 and 3 to keep the base model and examine gain by adding ESVEA parameters. Table S3. Based on models in table S2, ESVEA parameters were entered as dichotomised variable (above vs. below median). Table S4A. Based on models in table S2, LAVI/ a' was added as the strongest echocardiographic predictor of PAF.