Prediction of Transition to Chronic Atrial Fibrillation in Patients With Paroxysmal Atrial Fibrillation by Signal- Averaged Electrocardiography
A Prospective Study
Background It is well known that paroxysmal atrial fibrillation (PAF) often precedes the establishment of chronic atrial fibrillation (CAF). However, there have been no definite methods to predict the transition from PAF to CAF. The purpose of this report was to determine prospectively whether P-wave–triggered signal-averaged ECG (P-SAE) is useful for the prediction of the transition to CAF in patients with PAF.
Methods and Results One hundred twenty-two consecutive patients with PAF were prospectively followed after P-SAE, echocardiography, and 24-hour Holter monitoring at study entry. The duration (Ad) and root-mean-square voltage for the last 30 ms (LP30) of the filtered P wave were measured in P-SAE. The abnormality of P-SAE for the prediction of transition to CAF was defined as Ad ≥145 ms and LP30 <3.0 μV. Twenty-three (19%; group 1) of the patients had the abnormality of P-SAE, whereas the others (group 2) did not. During the follow-up period (mean, 26±12 months), 10 patients (43%) in group 1 acquired CAF, whereas the transition to CAF was observed in only 4 patients (4%) in group 2. Kaplan-Meier analysis revealed that the transition to CAF was significantly observed more often in group 1 than in group 2 (log-rank test, P<.0001). The Cox proportional hazards regression model identified that the variables most significantly associated with the transition to CAF were Ad (χ2=8.6, P=.003) and LP30 (χ2=5.1, P=.02), although significant differences in the left atrial dimension (40.8±5.3 versus 37.3±5.5 mm, P<.01) and the number of atrial premature contractions (3641±4524 versus 1489±2895 beats/d, P<.05) were observed between groups 1 and 2.
Conclusions These results indicate that P-SAE could be useful to identify patients at risk for the transition from PAF to CAF.
Paroxysmal atrial fibrillation is well known to often precede the establishment of CAF.1 2 3 4 5 In addition, some studies5 6 7 8 9 10 have revealed that the mortality and risk of thromboembolism are higher in patients with CAF than in those with PAF. Subsequently, it is of clinical importance to detect the risk of atrial fibrillation in patients with PAF.
There have been no definite methods to predict the transition from PAF to CAF, so the purpose of this study was to determine prospectively whether the transition from PAF to CAF could be predicted by the use of P-SAE, which we recently reported to be useful to identify the patients at risk for PAF during sinus rhythm.11 12
One hundred forty-nine consecutive outpatients with PAF who were underwent P-SAE between October 1992 and March 1995 in Osaka Prefectural Hospital were screened for this prospective study. PAF was considered present when both atrial fibrillation and sinus rhythm were documented on ECGs at <6 months before the undergoing P-SAE. Sixteen of 149 patients were excluded because (1) the mean noise level was ≥1 μV in the composite lead of P-SAE (5 patients), (2) it was difficult to determine the end of the filtered P wave because of too short PQ intervals (9 patients), or (3) the patient was taking an antiarrhythmic agent that could affect the results of P-SAE (2 patients). In addition, 11 patients who did not give consent to the follow-up period were excluded from a prospective analysis. A total of 122 patients were enrolled in this study.
The study patients were followed for 26±12 months (range, 6 to 38 months) after the entry, defined as the time of P-SAE. The interval from the first symptomatic episode, such as palpitation, to entry was 14±21 months in patients with symptomatic PAF attacks. The mean age of 122 patients was 61±12 years at entry. There were 77 men and 45 women. Fifty patients had no organic heart disease, and the remaining 72 patients had organic heart disease or another disease that possibly caused PAF (30 patients with ischemic heart disease, 14 with hypertension or hypertensive heart disease, 16 with valvular heart disease, 9 with pericardial or myocardial disease, 2 with congenital heart disease, and 1 with hyperthyroidism).
P-SAE was performed in all patients at the entry. The methodology of P-SAE recording and analysis has been described previously.11
The P-SAE was recorded from a modified X, Y, and Z lead system using the VCM-3000 (Fukuda Densi, Ltd). All of the digital data were stored on a floppy disk. The standard I lead was used as the X lead, the aVF lead was used as the Y lead, and the precordial V1 lead was used as the Z lead. The signal from each lead was amplified up to 5 μV/cm, passed through a low-pass filter of 300 Hz (slope, 12 dB/oct) and a high-pass filter of 40 Hz (slope, 18 dB/oct), and then converted from analog to digital data to a 12-bit accuracy at the sampling rate of 1 kHz.
A specially filtered P wave derived from the selected dominant sinus P wave of the standard II or V1 lead served as a reference signal for all processing. The specially filtered P wave was obtained with a band-pass filter of 10 to 30 Hz. The method of P-wave–triggered signal-averaging mainly involved three processes: (1) preparation of the template, (2) template matching, and (3) fine adjustment of the trigger point.
Preparation of Template
The template of the filtered P wave was made from the selected sinus P wave. The voltage was measured every 2 ms (31 points) from 30 ms before to 30 ms after the first peak of the specially filtered P wave; then, the specially filtered P wave was standardized so the total amplitude in this interval could be 250. We determined this wave in this window of the standardized filtered P wave as the template.
Template Matching (by a Difference Method)
A peak voltage of each specially filtered P wave was compared with the peak voltage of the template. Whenever the peak voltage of each specially filtered P wave was <70% of the template, the P wave was automatically rejected. Subsequently, the difference between each standardized P wave and the template was measured every 2 ms from 30 ms before to 30 ms after the adjusted trigger point of the specially filtered P wave. Whenever the sum [Σ ‖ X(template)i−Xi ‖ (1≤i≤31)] of the differences in 31 of these points in this window was >1024, the P wave was also automatically rejected; the number 1024 was determined in a preliminary experimental study so signal-averaging could be more precisely and quickly performed. If the number was <1024, it took longer to perform signal-averaging. Conversely, if the number was >1024, there was the fear that signals other than sinus P wave might be included.
Fine Adjustment of Trigger Point
Although the trigger point of signal-averaging had been determined around the first peak of the specially filtered P wave in advance, we finely adjusted the point in the following to make the triggering jitter least. Template matching was repeated with a fiducial point every 1 ms around the first peak of the specially filtered P wave. The trigger point was adjusted so the sum of differences in this window between each standardized P wave and the template might be minimized. The signals of 200 beats, which had already been filtered, were averaged on this trigger point within the specially filtered P wave. If the noise level remained >1 μV even after the averaging of 200 beats, averaging was continued until the peak noise level was reduced to <1 μV. The filtered signals for the X, Y, and Z leads were combined into a spatial magnitude: (X2+Y2+Z2)1/2. The onset and offset of the filtered P wave were defined as signals during the interval when signals show a persistent level of 1 μV. Ad and LP10, LP20, and LP30 for the last 10, 20, and 30 ms of filtered P wave were measured in the vector magnitude.
In our previous retrospective study,13 the SAE variables had been scrutinized for specific criteria indicating a risk of CAF in patients with PAF. The criteria had been defined as Ad ≥145 ms and LP30 <3.0 μV; therefore, these criteria were used in this study to predict the risk for CAF.
Echocardiography was also performed in all patients at entry (Toshiba SSH-160A recorder equipped with 2.5- or 3.5-MHz transducers). The standard technique14 was used for sizing of the left ventricle. Left ventricular dimensions were measured at end diastole, recognized for the peak of the R wave of the ECG, and at end systole, just below the mitral leaflets through the standard left parasternal window. Transducer position was aided by the two-dimensional echo mode, and the left ventricular ejection fraction was calculated according to Gibson’s method. Furthermore, the left atrial dimension was measured as a distance from the leading edge of the posterior aortic wall to the leading edge of the posterior left atrial wall at end systole.
In 82 of 122 patients, 24-hour ambulatory Holter ECG recordings were also obtained at entry. Analysis was performed using a Marquette Electronics 8000 Holter monitoring system, and the number of atrial premature contractions was counted for 24 hours.
Follow-up and Definition of CAF
All patients were followed up at least every month and examined by ECG or portable ECG monitoring to observe the cardiac rhythm. CAF was defined as atrial fibrillation sustained for ≥6 months. The results of P-SAE were blinded to the primary physicians taking care of the patients and did not in any way influence therapeutic decision. The use of antiarrhythmic agents (type and dosage) after the entry was left to the discretion of the primary physicians.
Data are presented as mean±SD. Statistical analysis was performed using Student’s t test to compare patients with and without the abnormality of P-SAE. The event (CAF)-free rates in patients with and without the abnormality of P-SAE were calculated using the Kaplan-Meier method, and the difference between them was detected using the log-rank test. The determination of the prognostic significance of abnormal results on P-SAE, echocardiography, and Holter monitoring was explored by survival analysis based on the Cox proportional hazards regression model. The level of significance was determined at a value of P=.05.
Clinical Characteristics in Patients With and Without the Abnormality of P-SAE
One hundred twenty-two patients were classified into one of two groups at study entry on the basis of the criteria for prediction of the risk of CAF in P-SAE: Ad ≥145 ms and LP30 <3.0 μV. Twenty-three patients had the abnormality of P-SAE (group 1), whereas the other 99 patients (group 2) did not. Fig 1⇓ shows the representative tracings of the P-SAE in patients in groups 1 and 2. Of note, the duration of the filtered P wave was longer in a patient in group 1 than in a patient in group 2. Furthermore, the terminal portion of the filtered P wave was lower in amplitude in group 1 than in group 2. Ad and LP30 were 161±17 ms and 2.2±0.4 μV in group 1 and 136±12 ms and 4.2±1.8 μV in group 2, respectively. Although there were no significant differences in sex, age, presence of organic heart disease, follow-up period, and use of class Ia or Ic antiarrhythmic agents, left atrial dimension was significant longer (40.8±5.3 versus 37.3±5.5 mm, P<.01) and the number of atrial premature beats was greater (3641±4524 versus 1488±2895 beats/d, P<.05) in group 1 than in group 2.
Prediction of Transition to CAF by P-SAE
There was no difference in follow-up period between groups 1 and 2 (25±13 versus 26±11 months). Ten patients (43%) in group 1 acquired CAF, whereas the transition to CAF was observed in only 4 patients (4%) in group 2. Patients with the abnormality of P-SAE had an 11-fold risk of transition to CAF. Fig 2⇓ shows the event (CAF)-free rate curve according to Kaplan-Meier analysis. In group 1, the CAF-free rate was 89% for 1 year and 38% for 3 years; in group 2, the CAF-free rate was 94% for 3 years. The transition to CAF in group 1 was significantly more frequently observed than in group 2 (P<.0001). Accordingly, abnormal findings of P-SAE gave a sensitivity of 71%, a specificity of 88%, a positive predictive value of 43%, and a negative predictive value of 96% for the prediction of transition to CAF.
The stepwise Cox survivorship analysis was used with regression covariates to determine the prognostic power of clinical variables (age, sex, presence of organic heart disease, and use of any antiarrhythmic agents) and noninvasive variables in P-SAE, echocardiography, and 24-hour Holter monitoring. Ad (χ2=8.6, P=.003), LP30 (χ2=5.1, P=.02), and age (χ2=4.9, P=.03) had the most significant relation to the transition to CAF, although Ad correlated with LP30 (r=−.374, P=.0001).
Comparison Between Patients With and Without the Transition to CAF
During the follow-up period (26±12 months; range, 6 to 38 months), the transition from PAF to CAF was observed in 14 of 122 patients (11%) (CAF group), and the other 108 patients continued to have PAF (PAF group). The clinical characteristics of the CAF and PAF groups are given in Table 2⇓. There were no significant differences in sex, age, and presence of organic heart disease between the CAF and PAF groups. Ad in the CAF group was significantly longer than that in the PAF group (162.7±19.8 versus 137.8±13.4 ms, P<.0001). Furthermore, there were significant differences in LP10, LP20, and LP30 between the two groups (LP10, 1.5±0.3 versus 1.9±0.7 μV, P<.05; LP20, 1.9±0.7 versus 2.6±1.1 μV, P<.05; and LP30, 2.5±0.8 versus 4.0±1.8 μV, P<.005). Echocardiographically, left atrial dimension in the CAF group was significantly larger than that in the PAF group (40.5±5.1 versus 36.4±5.2 mm, P<.05), although no significant differences in left ventricular dimensions measured at end diastole and end systole and ejection fraction were observed between the two groups. In 24-hour Holter monitoring, the number of atrial premature contractions for 24 hours in the CAF group (n=8) tended to be more than those in the PAF group (n=74) (3457±3848 versus 1741±3283 beats/d, P=.17).
Atrial fibrillation is a very common arrhythmia, with an incidence of 0.4% in the general population.15 It has been reported that in patients with electrocardiographically verified atrial fibrillation, <40% were paroxysmal.3 10 16 In addition, some investigators reported that PAF often developed into a permanent form for a period of >1 year with a rate of 5% to 33%1 2 3 4 5 and that the mortality and risk of thromboembolism were higher in patients with CAF than in those with PAF. Therefore, the risk of transition to CAF in patients with PAF should be taken into account in the treatment of PAF. However, no definite methods to predict the transition from PAF to CAF have been proposed. The present study indicates that the P-SAE could be useful to detect patients at risk for the transition to CAF; in particular, the duration and root-mean-square voltage for the last 30 ms of the filtered P wave would be potent markers to predict the transition from PAF to CAF.
Abnormality of P-SAE for Prediction of Transition to CAF
In our previous study,13 the signal-averaged ECG variables were scrutinized for specific criteria indicating a risk of CAF in patients with PAF. The P-SAE values of 14 patients who had accomplished CAF after undergoing P-SAE at the time of diagnosis of PAF were retrospectively compared with those of 68 patients who remained in PAF. Ad was significantly longer and the terminal portions of filtered P wave (LP10, LP20, and LP30) were significantly lower in 14 patients who had the transition to CAF than in the 68 patients who remained in PAF. When the criterion of Ad ≥145 ms was used, the sensitivity, specificity, and predictive accuracy values were 86%, 71%, and 80%, respectively; when the criterion of LP30 <3.0 μV was used, the sensitivity, specificity, and predictive accuracy values were 79%, 71%, and 72%, respectively. When the criteria of Ad ≥145 ms and LP30 <3.0 μV were combined, the values became 71%, 91%, and 89%, respectively. Therefore, the criteria of Ad ≥145 ms and LP30 <3.0 μV were used in this study to predict the risk for CAF.
Prediction of Transition to CAF
To our knowledge, there have been no published data regarding the methodology of detection of the risk for CAF. Takahashi et al3 reported that patients with the establishment of atrial fibrillation had more frequent and longer atrial fibrillation attacks than did those in whom the attacks remained paroxysmal. We previously reported17 that the filtered P-wave duration in patients with more frequent attacks (once a month or more often) was significantly longer than that in those with less frequent attacks (140.2±13.6 versus 130.8±11.1 ms), whereas there was no significant relation between the duration of PAF attacks and filtered P wave. In this study, 107 patients had symptomatic PAF attacks when the atrial fibrillation was documented in our hospital. However, PAF could not be documented in some patients when they had symptoms such as palpitation. Therefore, we investigated 57 patients who had documented PAF at least twice with symptom after entry. Four patients in group 1 and 23 patients in group 2 had a high frequency (once a month or more often) of PAF attacks at entry. On the other hand, 5 patients in group 1 and 26 patients in group 2 had longer (≥2 hours) PAF attacks. There were no significant differences in frequency (χ2=1.9, P=.17) and duration (χ2=1.7, P=.19) of PAF attacks between the two groups in this study.
The Framingham Study reported18 that left ventricular function was one of determinants of CAF. In the present study, however, there was no significant difference in left ventricular function measured by echocardiography between the CAF and PAF groups. The difference between the Framingham Study and the present study could be due to patient characteristics. In the present study, only outpatients who had maintained cardiac function relatively well were studied; most of the study patients had normal left ventricular function. Therefore, it is difficult to determine whether left ventricular function could be an important risk factor for CAF.
Pathophysiology of P-Wave–Duration Prolongation
Some investigators reported that patients with a history of atrial fibrillation had long signal-averaged P waves. Stafford et al19 tried to quantify differences in the fine morphology of P waves in a group of 9 patients with PAF versus 15 control subjects. They found signal-averaged P-wave duration was significantly increased in patients with PAF. Similarly, Guidera and Steinberg18 reported the filtered P-wave duration was longer in patients with a history of atrial fibrillation than in age- and disease-matched control subjects. We succeeded in detecting patients at risk for PAF during sinus rhythm by using P-SAE11 and discovered that filtered P-wave duration was significantly longer in patients with PAF than in control subjects. In this study, we observed that the patients with PAF who eventually developed CAF had further prolonged filtered P-wave duration compared with the patients who did not develop CAF. These results suggest that the conduction in atrium might be more severely disturbed in patients with CAF than in those with PAF, although this is highly speculative.
Clinical Implication of Prediction of Transition to CAF
It was clinically significant to predict the transition to CAF in patients with PAF because of the two following viewpoints. First, it was reported that mortality was higher in patients with CAF than in those with PAF regardless of the presence of underlying heart disease.7 Second, it was reported5 6 8 9 10 that the risk of thromboembolism was higher in patients with CAF than in those with PAF. In this study, 2 patients (14%) in the CAF group had systemic embolism during the follow-up period, whereas 2 patients (2%) in the PAF group had it; patients in the CAF group had a 7-fold risk of embolism compared with patients in the PAF group. Therefore, the patients with PAF who were at risk of CAF should have more frequent surveillance and possibly a more aggressive approach to the treatment of thromboembolism and their underlying heart diseases or a prophylactic intervention for atrial fibrillation.
First, because a PQ interval that is too short may mask the true end of filtered P wave in the initial portion of QRS complex in P-SAE, the end point of the filtered P wave could not be decided due to the overlap. However, if the filtered P wave intersected the QRS complex, we could analyze it as a tentative P wave from the onset of the P wave to the onset of the QRS complex. In this study, all 9 patients with a short PQ interval maintained PAF without a transition to the chronic stage. They would become classified into group 2 because of their relatively short P-wave duration. For these 9 patients, specificity would improve to 89%. Second, in this study, drug therapy was not controlled, so we could not precisely determine the influence of antiarrhythmic drugs in the transition to CAF. However, the primary physicians did not know the results of P-SAE, and there were no significant differences in the use of antiarrhythmic agents between the patients with and those without the abnormality of P-SAE. Third, in most of the patients, antiarrhythmic agents were administered to prevent PAF attacks after performance of control P-SAE on entry. This is why we were unable to analyze the serial change of P-SAE until the acquisition of CAF; the effect of antiarrhythmic agents on P-SAE was widely variable. However, the extent of the change in P-SAE due to the drugs could possibly reflect the risk of the transition from PAF to CAF. Further studies will be needed the resolve this matter. Fourth, the positive predictive value was low (43%); false-positives might be reduced by further studies that include optimal filtering. Fifth, in this study, we studied only consecutive outpatients who had maintained cardiac function, unlike inpatients. On this point, we can not overlook the bias in patient characteristics; therefore, further studies are needed to investigate patients with PAF who have clinical characteristics that differ from those of our patients.
In this study, we reported that patients with the abnormality of P-SAE had an 11-fold risk of the transition to CAF in comparison with those without the abnormality. These findings suggest that P-SAE was useful to predict the transition to CAF in patients with PAF.
Selected Abbreviations and Acronyms
|Ad||=||duration of filtered P wave|
|CAF||=||chronic atrial fibrillation|
|LP10, LP20, and LP30, respectively||=||root-mean-square voltage for the last 10, 20, and 30 ms of filtered P-wave, respectively|
|P-SAE||=||P-wave–triggered signal-averaged ECG|
|PAF||=||paroxysmal atrial fibrillation|
Reprint requests to Masatake Fukunami, MD, Division of Cardiology, Osaka Prefectural Hospital, 3-1-56 Mandai-Higashi, Sumiyoshi-ku, Osaka 558, Japan.
- Received January 21, 1997.
- Revision received May 12, 1997.
- Accepted May 28, 1997.
- Copyright © 1997 by American Heart Association
Peterson P, Godtfredsen J. Embolic complications in paroxysmal atrial fibrillation. Stroke. 1986;622-626.
Treseder AS, Sastry BSD, Thomas TPL, Yates MA, Pathy MS. Atrial fibrillation and stroke in elderly hospitalized patients. Age Ageing. 15:89-92,1986.
Godtfredsen J. Atrial Fibrillation: Etiology, Course and Prognosis: A Follow-up Study of 1212 Cases. Copenhagen, Denmark: Munksgaard; 1975.
Fukunami M, Yamada T, Ohmori M, Kumagai K, Umemoto K, Sakai A, Kondoh N, Minamino T, Hoki N. Detection of patients at risk for paroxysmal atrial fibrillation during sinus rhythm by P wave–triggered signal-averaged electrocardiogram. Circulation. 1991;83:162-169.
Abe Y, Fukunami M, Ohmori M, Kumagai K, Yamada T, Nishikawa N, Hoki N. Prediction of transition from paroxysmal atrial fibrillation by P wave–triggered signal-averaged electrocardiogram. Circulation. 1993;88(suppl II):II-312. Abstract.
Feigenbaum H, ed. Echocardiography. Philadelphia, Pa: Lea & Febiger; 1981.
Ostrander LD, Brandt RL, Kjelsberg MO, Epstein FH. Electrocardiographic findings among the adult population of a total natural community, Tecumseh, Michigan. Circulation. 1965;31:888-897.
Iwakura K, Abe Y, Ohmori M, Yamada T, Kondoh N, Minamino T, Tsujimura E, Fukunami M. Relationship between frequency and duration of paroxysmal atrial fibrillation attacks and atrial late potential. Circulation. 1992;86(suppl I):I-130. Abstract.