Electrocardiographic predictors of atrial fibrillation

Electrophysiology

Electrocardiographic predictors of atrial fibrillation Marco V. Perez, MD, a Frederick E. Dewey, MD, a Rachel Marcus, MD, a Euan A. Ashley, MRCP, DPhil, MD, a Amin A. Al-Ahmad, MD, a Paul J. Wang, MD, a and Victor F. Froelicher, MD, FACC a,b Stanford, CA

Background Atrial fibrillation (AF) is the most prevalent arrhythmia in the United States and accounts for more than 750,000 strokes per year. Noninvasive predictors of AF may help identify patients at risk of developing AF. Our objective was to identify the electrocardiographic characteristics associated with onset of AF. Methods This was a retrospective cohort analysis of 42,751 patients with electrocardiograms (ECGs) ordered by physician's discretion and analyzed using a computerized system. The population was followed for detection of AF on subsequent ECGs. Cox proportional hazard regression analysis was performed to test the association between these ECG characteristics and development of AF. Results For a mean follow-up of 5.3 years, 1,050 (2.4%) patients were found to have AF on subsequent ECG recordings. Several ECG characteristics, such as P-wave dispersion (the difference between the widest and narrowest P waves), premature atrial contractions, and an abnormal P axis, were predictive of AF with hazard ratio of approximately 2 after correcting for age and sex. P-wave index, the SD of P-wave duration across all leads, was one of the strongest predictors of AF with a concordance index of 0.62 and a hazard ratio of 2.7 (95% CI 2.1-3.3) for a P-wave index N35. These were among the several independently predictive markers identified on multivariate analysis. Conclusions

Several ECG markers are independently predictive of future onset of AF. The P index, a measurement of disorganized atrial depolarization, is one of the strongest predictors of AF. The ECG contains valuable prognostic information that can identify patients at risk of AF. (Am Heart J 2009;158:622-8.)

Atrial fibrillation (AF) is a cardiac arrhythmia characterized by disorganized atrial electrical activity leading to loss of effective contraction. Atrial fibrillation affects more than 2.2 million people in the United States,1 accounts for approximately 75,000 strokes per year,2 and is independently associated with a 1.5- to 1.9-fold increase risk of death.3 The primary goals of therapy for AF involve minimizing symptoms caused by AF through rhythm or rate control and lowering stroke risk with anticoagulant therapy. However, roughly 6.5% of patients who present with AFrelated strokes have no prior known history of AF.4 The ability to predict onset of AF may identify a group of patients whose stroke risk can be modified. Prediction of AF would also enable appropriate prophylactic therapy to be given to patients undergoing surgery. Approximately 10% to 40% of patients undergo-

From the aCardiovascular Medicine, Stanford University School of Medicine, Stanford, CA, and bCardiovascular Medicine, Palo Alto Veterans Affairs Health Care System, Stanford, CA. Submitted May 26, 2009; accepted August 6, 2009. Reprint requests: Marco V. Perez, MD, Falk CVRC, 300 Pasteur Drive, Stanford, CA 94305-5406. E-mail: [email protected] 0002-8703/$ - see front matter © 2009, Mosby, Inc. All rights reserved. doi:10.1016/j.ahj.2009.08.002

ing cardiothoracic surgery develop postoperative AF,5,6 which prolongs hospital stay and significantly augments resource use.7 Therapy with antiarrhythmics,8,9 steroids,10 and statins11 have been shown in randomized trials to help prevent this postoperative arrhythmia. Prediction of AF could help identify the patients who would benefit most from prophylactic use of these medical interventions. Several noninvasive predictors of AF have been identified. After age, left atrial size on echocardiography remains one of the strongest predictors of AF.12-14 Premature atrial contractions (PACs) on Holter monitoring have also been strongly associated with recurrence of AF.15,16 Studies of signal-averaged electrocardiograms (ECGs) have revealed characteristics that predict postoperative AF,17,18 progression from paroxysmal to permanent AF,19,20 or onset of AF in high-risk groups.21,22 The 12-lead resting ECG, however, remains the most frequently used study in the evaluation of patients for cardiovascular disease and, because of its relatively low cost, has the greatest potential to be used as a screening tool. Maximal P-wave duration (Pmax) and P-wave dispersion (Pdisp), defined as the difference between the Pmax and Pmin across the 12 ECG leads, have been consistently reported as predictors of postoperative AF,23 frequent episodes of AF paroxysms,24 and recurrence of AF after cardioversion.25-27 Most of these studies were

American Heart Journal Volume 158, Number 4

performed on a select group of patients known either to have an established history of AF or to be at high postoperative risk of AF. One recent population-based study identified Pmax and P-wave morphologies as very strong predictors of AF.28 However, these studies relied on manual, and often meticulous, measurements of the P waves. We sought to evaluate the computerized 12-lead resting ECG for its ability to predict AF in a large cohort of patients not necessarily known to be at high risk of AF. We introduce the P-wave index (Pindex), defined as the SD of P-wave duration across the 12 leads, and hypothesized that this and several other ECG markers would independently be predictive of subsequent conversion to AF.

Methods Study design The Palo Alto Veterans Affairs Health Care System uses a computerized ECG system (GE Marquette) to collect, store, and analyze ECGs. This system has been validated by both the US Food and Drug Administration and the European Community and is widely used across the world. The current study involved the retrospective analysis of 45,855 initial ECGs obtained between March 1987 and July 2000 that were ordered for usual clinical indications. The 3,104 patients found to be in AF on the initial ECG were excluded from this analysis. Age, gender, race, weight, and height of each patient were also recorded. No extramural funding was used to support this work. The authors are solely responsible for the design and conduct of this study, all study analyses, the drafting, and editing of the article and its final contents.

Perez et al 623

clinical indications. Determination of AF, both on initial ECG and on subsequent ECGs, was computerized as noted above. Patients were followed until July 2000, the onset of AF or death, whichever came first. Death was determined using the Veterans Affairs Health Care System electronic medical record or the California Health Department Service database. Study personnel blinded to ECG results determined the time of death. Review of the Veterans Affairs medical record provided a level of accuracy superior to death certificates.

Statistical analysis Bivariate associations between subsequent AF onset and various ECG criteria were tested via χ2 tests (dichotomous variables) and Student t tests (continuous variables). All variables were first tested for normality of distribution. Statistical significance was defined by P b .05. Cox proportional hazards regression analysis was performed to determine whether the ECG and demographic variables were predictive of subsequent development of AF. Hazard ratios (HRs) with 95% CI for several ECG characteristics were first calculated using a model adjusted for age and sex. The HRs for several ECG characteristics were then calculated using a larger multivariate model adjusted for age, sex, PR interval, PACs, abnormal P axis, prolonged P duration, P wave duration N35, LEA, PVCs, left bundle branch block (LBBB), and LVH. Proportional hazard assumptions were met as verified by plotting the log-negative-log of within-group freedom from AF. The Kaplan-Meier method was used to estimate the rates of onset of AF in subgroups of Pindex and RENN scores. SAS (v9.1; SAS, Cary, NC) and the “design” and “Hmisc” libraries in S-Plus version 7.0 (Insightful Corp, Seattle, WA) were used for statistical analysis.

Electrocardiogram analysis

Results

The recorded data on each ECG included the timing and voltages at each of the points of the PQRST complex of the basic 8 leads with derivation of the remaining 4 leads. The system was able to flag rhythm abnormalities, measure standard intervals, and perform waveform analysis to provide the basic electrocardiographic interpretations (GE 12 SL analysis program, www. gesystems.com). Standardized computerized ECG criteria as described by the GE 12-lead electrographic analysis program were used for the diagnosis of Q waves, ST changes, left atrial enlargement (LEA), right atrial enlargement (RAE), abnormal P axis, PAC, premature ventricular contraction (PVCs), and bundle branch blocks. From these measurements, the Romhilt-Estes29 criteria for left ventricular hypertrophy (LVH), spatial QRS-T angle,30 Cardiac Infarction Injury Scores (CIISs),31 Selvester scores,32 and the Resting ECG Neural Network (RENN) scores33 were calculated. Pdisp is the difference between the Pmax and Pmin, Pmean is the mean P-wave duration, and Pindex is the SD of the P-wave duration across the 12 ECG leads.

Baseline demographics and ECG characteristics There were a total of 42,751 patients with an initial ECG demonstrating absence of AF. These patients were followed for an average of 5.3 years and received a mean of 3.2 follow-up ECGs. Fifty-three percent of patients received 1 ECG, 15% had 2 ECGs, and 32% had more than 2 ECGs. The period between ECGs was, on average, 3.0 years. During this period, 1,050 (2.4%) patients were found to have AF on a subsequent ECG. Table I presents the baseline demographic information of patients who maintained sinus rhythm versus those who converted to AF. Those who converted to AF were older (67.5 vs 55.8 years, P b .0001) and were less likely to be female (3.6% vs 10.5%, P b .0001) or Hispanic (3.9% vs 12.7%, P b .0046) but were more likely to be black (8.0% vs 6.2%, P b .0001). There were no statistically significant differences in body mass index (BMI) between the 2 groups. No statistically significant differences in heart rate between the 2 groups were detected at baseline (Table II). Several ECG abnormalities, including PACs, Pindex, Pdisp, abnormal P axis, left atrial enlargement (LAE), LBBB, and prolonged QTc intervals, were found

Outcomes The primary outcome evaluated for this study was detection of AF on a subsequent ECG. The 42,751 patients found initially to be in sinus rhythm were followed for evidence of AF with a total of 137,167 follow-up ECG recordings spanning from March 1987 to July 2000. Follow-up ECGs were also obtained for usual

American Heart Journal October 2009

624 Perez et al

Table I. Baseline demographic information of patients who maintained sinus rhythm versus those who converted to AF Sinus maintained, n = 41 701

AF conversion, n = 1050

P

55.8 ± 15 10.5% (0.2) 1.74 ± 0.1 83.0 ± 18 27.3 ± 5.6 12.7% (0.2) 6.2% (0.1)

67.5 ± 10.5 3.6% (0.6) 1.75 ± 0.09 84.2 ± 17 27.5 ± 5.5 3.9% (0.6) 8.0% (0.9)

b.0001 b.0001 .0008 .05 .35 .0046 b.0001

Age (y) Female Height (m) Weight (kg) BMI (kg/m2) Hispanic Black

HR (CI)

Values represent mean ± SD or percentage (SE). P values are for differences between patients who maintained sinus rhythm and those who converted to AF. P b .05 was considered statistically significant.

Table II. Baseline ECG characteristics of patients who maintained sinus rhythm versus those who converted to AF Sinus maintained, AF conversion, n = 41 701 n = 1050 Heart rate (beat/min) PR interval (ms) Atrial markers PAC Abnormal P axis LEA Pmean (ms) Pmax (ms) Pmin (ms) Pdisp (ms) Pindex (ms) RAE Ventricular markers PVC Left axis deviation Left bundle branch block Right bundle branch block QRS N110 ms Pathologic Q waves QTc N450 LVH (RomhiltEstes N 3) Advanced markers QRST spatial angle N100 CIIS N30 Selvester score N6 Simple score N3 RENN

Table III. Hazard ratios for developing AF on follow-up ECG using Cox regression models adjusted for age and sex

P

74 ± 14.6

73.8 ± 15

.65

161.1 ± 33

176 ± 44

b.0001

1.9% (0.07) 4.0% (0.1) 4.1% (0.1) 95.9 ± 13 106.2 ± 14 47.8 ± 17 58.4 ± 20 20.9 ± 8.0 0.6% (0.04)

7.6% (0.8) 9.8% (1.0) 9.3% (0.9) 97.6 ± 17 113 ± 20 48.8 ± 16 63.8 ± 23 24.2 ± 9.5 1.1% (0.3)

b.0001 b.0001 b.0001 b.0001 b.0001 .07 b.0001 b.0001 .05

4.3% (0.1) 8.6% (0.1)

10.3% (0.9) 17% (1.2)

b.0001 b.0001

1.1% (0.05)

3.8% (0.6)

b.0001

3.2% (0.09)

7.5% (0.8)

b.0001

10.6% (0.2) 12.2% (0.2)

25% (1.3) 20% (1.3)

b.0001 b.0001

19.6% (0.2) 5.3% (0.1)

36% (1.5) 11.5% (1.0)

b.0001 b.0001

8.3 (0.14)

20.8% (1.2)

b.0001

5.1% (0.1) 7.6% (0.13)

12.8% (1.0) 15.5% (1.1)

b.0001 b.0001

1.5% (0.06) 0.42 ± 0.3

11.5% (1.0) 0.66 ± 0.3

b.0001 b.0001

Values represent mean ± SD or percentage (SE). P values are for differences between patients who maintained sinus rhythm and those who converted to AF. P b .05 was considered statistically significant.

Demographics Male sex⁎ Height (per cm) BMI (per kg/m2) Hispanic Black Basic markers Heart rate PR interval N200 ms Atrial markers PAC Abnormal P axis LEA Pmax N120 ms Pdisp N80 Pindex N30 Pindex N35 RAE Ventricular markers PVC Left axis deviation Left bundle branch block Right bundle branch block QRS N110 ms Q waves present QTc N450 Pathologic Q waves LVH (Romhilt-Estes N3) Advanced markers QRST spatial angle N100 CIIS N30 Selvester score N6 Simple score N3 RENN high risk

P

(3.0-5.7) (1.01-1.03) (1.01-1.04) (0.6-1.1) (0.7-1.1)

b.0001 b.0001 .0003 .14 .29

1.004 (1.0-1.009) 1.8 (1.5-2.07)

.07 b.0001

4.1 1.02 1.03 0.8 0.9

2.1 2.1 1.9 1.9 1.95 2.1 2.7 1.6

(1.7-2.7) (1.7-2.6) (1.5-2.4) (1.7-2.2) (1.7-2.3) (1.8-2.4) (2.1-3.3) (0.9-2.7)

b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 .13

1.7 1.3 2.2 1.5 1.9 1.4 1.7 1.4 1.8

(1.3-2.1) (1.1-1.6) (1.6-3.1) (1.2-1.8) (1.7-2.2) (1.2-1.7) (1.5-2.0) (1.2-1.7) (1.5-2.2)

b.0001 .0009 b.0001 .0012 b.0001 b.0001 b.0001 b.0001 b.0001

1.9 1.9 1.5 2.2 2.1

(1.7-2.3) (1.6-2.3) (1.3-1.8) (1.6-3.0) (1.7-2.5)

b.0001 b.0001 b.0001 b.0001 b.0001

RENN high risk, RENN score N0.8. P b .05 was considered statistically significant. ⁎ The HR for sex was adjusted for age only; all other HRs were adjusted for sex and age.

more frequently in the patients with subsequent AF. The PR intervals were longer (176 vs 161 milliseconds, P b .0001), and Pindex was more pronounced (24.2 vs 20.9, P b .0001) in the AF conversion group. In addition, several advanced electrocardiographic markers that have traditionally been used to predict cardiovascular mortality were statistically significantly elevated in the group of patients who were to subsequently develop AF, including the RENN score (0.42 vs 0.66, P b .0001).

Cox regression analysis After adjusting for age and sex in the proportional hazards models, several ECG findings were individually evaluated for associations with subsequent development of AF (Table III). Heart rate was not a determinant of subsequent AF conversion; however, a prolonged PR interval, defined as greater than 200 milliseconds, was predictive of AF (HR 1.8, 95% CI 1.5-2.07). Most atrial markers studied, such as PACs and an abnormal P axis,

American Heart Journal Volume 158, Number 4

Perez et al 625

Table IV. Concordance indexes for the continuous ECG variables evaluated for prediction of AF Continuous ECG characteristic

C-index

Table V. Hazard ratios for developing AF on follow-up ECG using a multivariate Cox regression model adjusted for all factors in the table HR (CI)

Heart rate PR interval Pmax Pmean Pdisp Pindex QRS duration QTc RENN

0.50 0.62 0.61 0.52 0.60 0.62 0.60 0.62 0.76

Age (per year) Male sex PR interval N200 ms PAC Abnormal P axis Pmax N120 ms Pindex N35 ms LEA PVC LBBB LVH

1.07 4.4 1.3 2.1 1.9 1.6 1.7 1.5 1.5 1.7 1.3

(1.06-1.08) (3.1-6.5) (1.1-1.6) (1.6-2.7) (1.6-2.4) (1.3-1.8) (1.3-2.1) (1.1-2.0) (1.2-1.9) (1.2-2.5) (1.0-1.7)

P b.0001 b.0001 .003 b.0001 b.0001 b.0001 b.0001 .011 .0002 .004 .046

The HRs reported have each been adjusted for every other variable in the table. P b .05 was considered statistically significant.

Figure 1

Cox proportional HRs of developing AF for Pindex N35 adjusted for age and sex at different age intervals. The HRs reported were for Pindex N35 compared with Pindex b35 in each individual age category. Error bars represent the 95% CI of these HRs. The only 95% lower limit CI that crosses 1.0 was in the 80 to 90 intervals. Age groups are expressed in years.

conferred an approximately 2-fold increase in the hazard rate for developing AF. The strongest atrial ECG characteristic evaluated for prediction of AF was Pindex N35, generating an HR of 2.7 (95% CI 2.1-3.3). When evaluated as a continuous variable, the concordance index (C-index) for Pindex (0.62) was higher than those of Pdisp (0.60) or Pmax (0.61) (Table IV). Interestingly, the predictive power of Pindex N35 was highest at both extremes of age accounting for an HR of 5.3 (95% CI 2.34.7) at age b60 years and an HR of 7.4 (95% CI 1.6-35) at age N90 years (Figure 1). Several ventricular ECG markers, such as LBBB, LVH, and a QTc interval greater than 450 also predicted development of AF with statistical significance (Table III). In addition to these individual markers,

several scores traditionally used to predict cardiovascular mortality, such as the RENN score (HR 2.1, 95% CI 1.8-2.5), also demonstrated their predictive power. The RENN score, a value that uses an artificial neural network to predict cardiovascular mortality, was the continuous ECG variable with the highest predictive value, at a C-index of 0.76. A multivariate model was then created using the most clinically relevant and strongest demographic and electrocardiographic predictors. Table V demonstrates the HR generated from this model with each value adjusted for all other variables in the table. Most notably, each of these markers contributed independently to the risk of future development of AF. In the multivariate model, presence of PACs, abnormal P axis, and Pindex N35 were the strongest independent ECG predictors. The advanced ECG markers were then individually and separately added to this multivariate model, and each marker retained an independent ability to predict AF. A high RENN score (N0.8) independently predicted onset of AF with an adjusted HR of 1.8 (95% CI 1.6-2.1) when compared with a low or medium score.

Kaplan-Meier survival analysis To further illustrate the association between Pindex on subsequent development of AF, we generated a KaplanMeier plot of freedom from AF for increasing measurements of Pindex (Figure 2). There is a clear separation of the plots with a log-rank statistic of P b .0001. Similarly, increasing RENN score risk categories also showed a clear separation of the plots on a Kaplan-Meier analysis with a log-rank statistic of P b .0001 (Figure 3).

Discussion Atrial fibrillation has been growing at a pace rapid enough to be labeled an epidemic.34 Because the population continues to grow and age, the prevalence

American Heart Journal October 2009

626 Perez et al

Figure 2

Kaplan-Meier plot of cumulative freedom from AF by categories of Pindex. Log-rank statistic P b .0001.

of AF and its heavy burden on health care will continue to surge. Novel therapies such as better-tolerated anticoagulants, more selective antiarrhythmics, and catheter interventions are evolving and will make treatment of AF less challenging in the future. As these therapies for AF become safer and more affordable, screening for AF will gain importance. Currently, many patients present with a stroke as the first manifestation of AF.4 The ability to noninvasively identify the patients at high risk of AF-related strokes and begin prophylactic anticoagulation will prove invaluable. These noninvasive markers of AF may also help monitor response to antiarrhythmic therapy or catheter-based interventions and identify patients who are likely to have progression of atrial disease. Although there are alternative noninvasive tests that may aid in predicting AF, such as echocardiography, signal-averaged ECGs, and Holter monitoring, the resting ECG remains the least expensive and most routinely used test for the evaluation of cardiovascular disease. As opposed to manual measurements, the computerized ECG provides quantifiable and easily reproducible measurements. Because of the large size of our study, most of the variables evaluated had a statistically significant association with future onset of AF. However, only a handful of these variables also demonstrated a clinically important association, with HRs above 2.0; only some of these proved to be independent predictors of this arrhythmia. We have confirmed prior observations that Pdisp and other measures of atrial electrical activity are predictive of AF (Table III). However, one of the strongest predictors of AF was Pindex N35. Prolonged P-wave duration, which reflects the time it takes for the atria to depolarize, itself can be a marker of atrial disease. Differences in P-wave duration seen in different leads

Figure 3

Kaplan-Meier plot of cumulative freedom from AF by RENN class. RENN score (high risk = RENN N 0.8, mod risk = RENN 0.4-0.8, low risk = RENN b 0.4). Log-rank statistic P b .0001.

can be a function of either differences in conduction velocities in different areas of the atria or of marked asymmetry of the atria themselves. Pindex accounts for the differences in atrial conduction across different vectors and is a novel measurement that may more accurately represent the variability in atrial conduction, which in turn may better reflect the heterogeneity of diseased atria. The risk of AF in patients with a high Pindex was most pronounced at the extremes of age (Figure 1), which suggests that screening for AF may be of greater benefit in certain populations. Young patients with evidence of abnormal atrial conduction (high Pindex) may represent a group that is particularly prone to AF, whereas much older populations with signs of normal atrial conduction (normal Pindex) may already demonstrate protection against development of AF. The risk of AF with elevated Pindex measurements was independent of LAE, overall P duration, or an abnormal P axis. Each of these characteristics likely reflects a different aspect of atrial shape, myocardial conduction, and function. The observation that PACs are predictive of AF onset comes at no surprise given the role of PACs in triggering AF. It will be interesting in the near future to better characterize the origin of these PACs to determine the relative importance of the different atrial structures in contributing to the risk of AF. Although ventricular disease may lead to AF via the consequent dilation of the left atrium, many of the ventricular markers analyzed also contributed to the risk of AF independent of atrial abnormalities. Electrocardiogram markers of LVH have previously been reported to be risk factors of AF35; and regression of these markers has proven protective of AF.36 However, the observation that

American Heart Journal Volume 158, Number 4

LBBB and PVCs also contribute to the risk may be a reflection of more global conduction disturbances, such as a propensity for myocardial fibrosis, that may be modulating AF risk. In addition to the observation that individual markers on the ECG predict AF, we have shown that several risk scores such as the RENN score, which attempt to make a more generalized characterization of cardiovascular risk based on the ECG, can also predict AF (Table III). Although not specifically designed to predict AF, their association with future AF onset supports prior observations that patients at risk of cardiovascular death also have higher rates of AF. One of the next steps will be to create a risk assessment that accounts for all of the individual factors and the interactions between them. Although regression analyses have served well in the past to create these models, studies with neural networks33 demonstrate that methods that take into account interactions between predictive variables can outperform traditional statistical approaches. There are a few limitations to the current study worth noting. The reasons for referral to initial and follow-up ECG testing have not been thoroughly documented; however, these ECGs were ordered by physician discretion, which is comparable to a real-world setting. Of the patients with repeat ECGs, those with AF had a 3-fold increase in the number of ECGs performed. It is possible that more frequent monitoring may have increased the likelihood of detecting AF. However, because the number of repeat ECGs performed was at the discretion of the referring physician, it is likely that patients with symptoms of AF were more likely to have multiple ECGs taken. Importantly, the number of followup ECGs being performed was not determined by the ECG findings such as Pindex and RENN scores. In addition, the diagnosis of AF was based on computerized interpretation, which allowed for an unbiased and efficient screening of a large population; however, this will have resulted in some degree of misclassification. Because the determination of future AF is not influenced by the findings in the original ECGs, this misclassification would be unbiased and would only result in dilution of the effects reported. Another limitation not confined to this study is the intermittent and often asymptomatic nature of some forms of AF, which may lead to an underestimation of AF on follow-up. However, this is likely a nondifferential bias that would also dilute the effects of the characteristics under study. The ECGs were gathered from a predominantly male, white, veteran population. Further clinical information about the patient cohort, such as history of heart failure and echocardiographic markers, is not available. The findings need to be validated in a more generalized patient cohort with documentation of these baseline clinical characteristics.

Perez et al 627

Summary In summary, we retrospectively studied more than 40,000 patients who were followed for the development of AF with serial ECGs. We identified several independent electrocardiographic risk factors for the development of AF, including Pindex, a novel measurement that may better represent atrial heterogeneity. These risk factors can be used to identify patients at high risk of developing AF who may benefit from prophylactic anticoagulant or antiarrhythmic therapy to prevent strokes and hospitalizations. Further studies will need to be performed to assess the utility of a screening method using these variables.

References 1. Feinberg WM, Blackshear JL, Laupacis A, et al. Prevalence, age distribution, and gender of patients with atrial fibrillation. Analysis and implications. Arch Intern Med 1995;155:469-73. 2. Wolf PA, Mitchell JB, Baker CS, et al. Impact of atrial fibrillation on mortality, stroke, and medical costs. Arch Intern Med 1998;158: 229-34. 3. Benjamin EJ, Wolf PA, D'Agostino RB, et al. Impact of atrial fibrillation on the risk of death: the Framingham Heart Study. Circulation 1998;98:946-52. 4. Staszewski J. Atrial fibrillation characteristics in patients with ischaemic stroke. Kardiol Pol 2007;65:751-7. 5. Michelson EL, Morganroth J, MacVaugh III H. Postoperative arrhythmias after coronary artery and cardiac valvular surgery detected by long-term electrocardiographic monitoring. Am Heart J 1979;97:442-8. 6. Crosby LH, Pifalo WB, Woll KR, et al. Risk factors for atrial fibrillation after coronary artery bypass grafting. Am J Cardiol 1990;66: 1520-2. 7. Aranki SF, Shaw DP, Adams DH, et al. Predictors of atrial fibrillation after coronary artery surgery. Current trends and impact on hospital resources. Circulation 1996;94:390-7. 8. Daoud EG, Strickberger SA, Man KC, et al. Preoperative amiodarone as prophylaxis against atrial fibrillation after heart surgery. N Engl J Med 1997;337:1785-91. 9. Giri S, White CM, Dunn AB, et al. Oral amiodarone for prevention of atrial fibrillation after open heart surgery, the Atrial Fibrillation Suppression Trial (AFIST): a randomised placebo-controlled trial. Lancet 2001;357:830-6. 10. Halonen J, Halonen P, Jarvinen O, et al. Corticosteroids for the prevention of atrial fibrillation after cardiac surgery: a randomized controlled trial. Jama 2007;297:1562-7. 11. Patti G, Chello M, Candura D, et al. Randomized trial of atorvastatin for reduction of postoperative atrial fibrillation in patients undergoing cardiac surgery: results of the ARMYDA-3 (Atorvastatin for Reduction of MYocardial Dysrhythmia After cardiac surgery) study. Circulation 2006;114:1455-61. 12. Cui Q, Zhang W, Wang H, et al. Left and right atrial size and the occurrence predictors in patients with paroxysmal atrial fibrillation. Int J Cardiol 2007. 13. Tanabe K, Yamaguchi K, Tani T, et al. Left atrial volume: predictor of atrial fibrillation in patients with degenerative mitral regurgitation. J Heart Valve Dis 2007;16:8-12. 14. Bollmann A, Husser D, Steinert R, et al. Echocardiographic and electrocardiographic predictors for atrial fibrillation recurrence

628 Perez et al

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

following cardioversion. J Cardiovasc Electrophysiol 2003;14(10 Suppl):S162-S165. Varounis C, Dagres N, Maounis T, et al. Atrial premature complexes and heart rate have prognostic significance in 1-month atrial fibrillation recurrence after electrical cardioversion. Europace 2007; 9:633-7. Thong T, McNames J, Aboy M, et al. Prediction of paroxysmal atrial fibrillation by analysis of atrial premature complexes. IEEE Trans Biomed Eng 2004;51:561-9. Steinberg JS, Zelenkofske S, Wong SC, et al. Value of the P-wave signal-averaged ECG for predicting atrial fibrillation after cardiac surgery. Circulation 1993;88:2618-22. Hayashida N, Shojima T, Yokokura Y, et al. P-wave signal-averaged electrocardiogram for predicting atrial arrhythmia after cardiac surgery. Ann Thorac Surg 2005;79:859-64. Koide Y, Yotsukura M, Sakata K, et al. Investigation of the predictors of transition to persistent atrial fibrillation in patients with paroxysmal atrial fibrillation. Clin Cardiol 2002;25:69-75. Abe Y, Fukunami M, Yamada T, et al. Prediction of transition to chronic atrial fibrillation in patients with paroxysmal atrial fibrillation by signal-averaged electrocardiography: a prospective study. Circulation 1997;96:2612-6. Cecchi F, Montereggi A, Olivotto I, et al. Risk for atrial fibrillation in patients with hypertrophic cardiomyopathy assessed by signal averaged P wave duration. Heart 1997;78:44-9. Montereggi A, Marconi P, Olivotto I, et al. Signal-averaged P-wave duration and risk of paroxysmal atrial fibrillation in hyperthyroidism. Am J Cardiol 1996;77:266-9. Piechowiak M, Banach M, Ruta J, et al. Risk factors for atrial fibrillation in adult patients in long-term observation following surgical closure of atrial septal defect type II. Thorac Cardiovasc Surg 2006;54:259-63. Dilaveris PE, Gialafos EJ, Andrikopoulos GK, et al. Clinical and electrocardiographic predictors of recurrent atrial fibrillation. Pacing Clin Electrophysiol 2000;23:352-8. Boriani G, Diemberger I, Biffi M, et al. P wave dispersion and short-term vs. late atrial fibrillation recurrences after cardioversion. Int J Cardiol 2005;101:355-61.

American Heart Journal October 2009

26. Amasyali B, Kose S, Aytemir K, et al. P wave dispersion predicts recurrence of paroxysmal atrial fibrillation in patients with atrioventricular nodal reentrant tachycardia treated with radiofrequency catheter ablation. Ann Noninvasive Electrocardiol 2006;11:263-70. 27. Aytemir K, Amasyali B, Kose S, et al. Maximum P-wave duration and P-wave dispersion predict recurrence of paroxysmal atrial fibrillation in patients with Wolff-Parkinson-White syndrome after successful radiofrequency catheter ablation. J Interv Card Electrophysiol 2004; 11:21-7. 28. De Bacquer D, Willekens J, De Backer G. Long-term prognostic value of p-wave characteristics for the development of atrial fibrillation in subjects aged 55 to 74 years at baseline. Am J Cardiol 2007;100: 850-4. 29. Romhilt DW, Estes Jr EH. A point-score system for the ECG diagnosis of left ventricular hypertrophy. Am Heart J 1968;75:752-8. 30. Ball MF, Pipberger HV. The normal spatial QRS-T angle of the orthogonal vectorcardiogram. Am Heart J 1958;56:611-5. 31. Rautaharju PM, Warren JW, Jain U, et al. Cardiac Infarction Injury Score: an electrocardiographic coding scheme for ischemic heart disease. Circulation 1981;64:249-56. 32. Palmeri ST, Harrison DG, Cobb FR, et al. A QRS scoring system for assessing left ventricular function after myocardial infarction. N Engl J Med 1982;306:4-9. 33. Perez MV, Dewey FE, Tan SY, et al. Added value of a resting ECG neural network that predicts cardiovascular mortality. Ann Noninvasive Electrocardiol 2009;14:26-34. 34. Braunwald E. Shattuck lecture—cardiovascular medicine at the turn of the millennium: triumphs, concerns, and opportunities. N Engl J Med 1997;337:1360-9. 35. Wachtell K, Lehto M, Gerdts E, et al. Angiotensin II receptor blockade reduces new-onset atrial fibrillation and subsequent stroke compared to atenolol: the Losartan Intervention For End Point Reduction in Hypertension (LIFE) study. J Am Coll Cardiol 2005;45:712-9. 36. Okin PM, Wachtell K, Devereux RB, et al. Regression of electrocardiographic left ventricular hypertrophy and decreased incidence of new-onset atrial fibrillation in patients with hypertension. Jama 2006;296:1242-8.