Letter by Berenfeld and Jalife Regarding Article “Dominant Frequency of Atrial Fibrillation Correlates Poorly With Atrial Fibrillation Cycle Length”

To the Editor

We read with interest the recent article by Elvan et al in Circulation: Arrhythmia and Electrophysiology.¹ They compared the reliability of time-domain versus frequency-domain methods in determining the local activation rate in patients during atrial fibrillation (AF).² The study provided potentially important information regarding the analysis of unipolar versus bipolar recordings. Unfortunately, essential details regarding the methods used for both the time-domain and the frequency-domain analyses were not presented. For example, information on the noise and frequency rejection criteria used for determination of AF cycle length in the time domain was nowhere to be found. Also crucial information on the possible contributions of far and near fields was absent. Importantly, because the authors indicated that they performed manual editing for AF cycle length determination, a comparison between results obtained by at least 2 different operators would have been reassuring but was not included. As for the power spectral analysis used to determine the dominant frequency (DF), the approach of Elvan et al differed substantially from previous studies.^2–5 First, Elvan et al chose to analyze AF episodes that were about 3-fold longer than typically used by others. Second, it is unclear whether or not they detrended and rectified the signal, as was reported previously. Third, these authors chose to manipulate their signals in such a way that the filters used contributed significantly to distort the frequency content in their data relative to other studies. Fourth, to our knowledge, Elvan et al did not perform any regularity-based screening to reduce the burden of spurious DF results. Finally, the same authors elected to exclude from their analysis DFs they considered nonphysiological, which, unless demonstrated otherwise, probably biased their correlation results. The alternative and more rigorous approach would have been to determine the DF in the physiological range.

All of the above deficiencies in the article of Elvan et al make it impossible to reproduce and/or compare its results with those of previous publications. Moreover, Elvan et al do not correlate their data with any independent measure of activation rate or any intervention, which makes their study effectively inadequate when attempting to decide whether one or the other method is better.^3–5 In other words, the major conclusion drawn by Elvan et al that time-domain methods are more preferable than frequency-domain methods for the detection of high frequency sites is unjustified by the data presented.

Finally, the authors’ assertion that the multiple wavelet hypothesis is the most commonly accepted for the mechanism underlying AF is clearly outdated. Not only there is substantial evidence that paroxysmal AF in the majority of cases depends on the activity of high frequency sources at or near the pulmonary veins, but the actual presence of left-to-right AF cycle length gradients in the data of Elvan et al, as well as in many other previously published studies, makes it untenable to conclude that fibrillatory activity is purely random, as the multiple wavelet hypothesis would predict.