TfsNBGA

FtI

BdglaWq

Oc

Mu1mcc

RfimC

II(cfibCf

esaCs2

1

ARTICLE IN PRESS

he temporal variability of dominant frequency and complexractionated atrial electrograms constrains the validity ofequential mapping in human atrial fibrillationicole Habel,*§ Pierre Znojkiewicz, MD,*† Nathaniel Thompson, MD,*† Joachim G. Müller, MD,*†

ryan Mason, BA,* James Calame, RN, BSN,*† Susan Calame, RN,† Shruti Sharma, BS,‡

agan Mirchandani, PhD,‡ Deborah Janks, PhD,*§ Jason Bates, PhD,*§ Arshia Noori, MD,*†

ndreas Karnbach,*§ Daniel L. Lustgarten, MD, PhD,*†§ Burton E. Sobel, MD,*†§ Peter Spector, MD*†§

rom the *Department of Medicine at the University of Vermont College of Medicine and †Fletcher Allen Health Care,he ‡College of Engineering and Computer Science at the University of Vermont, and the §Cardiovascular Research

nstitute, Burlington, Vermont.

o1

CCetqs

KqD

AtClHd

(

ACKGROUND It has been proposed that sequential mapping ofominant frequency (DF) and complex fractionated atrial electro-rams (CFAE) can identify target sites for ablation of atrial fibril-ation (AF). These mapping strategies are valid only if DF and CFAEre temporally stable on the timescale of the mapping procedure.e postulate that DF and CFAE are temporally variable; conse-uently, sequential mapping can be misleading.

BJECTIVE To make prolonged spatially stable multielectrode re-ordings to assess the temporal stability of DF and CFAE.

ETHODS We recorded electrical activity for 5 minutes with these of a 64-electrode basket catheter placed in the left atrium of8 patients presenting for AF ablation. DF and CFAE were deter-ined off-line, and their temporal variability was quantified. Maps

reated from simultaneous versus sequentially acquired data wereompared.

ESULTS DF was temporally variable: the average temporal coef-cient of variation was 22.7% � 5.4%. DF sites were transient,eeting criteria for only 22.1 seconds out of 5 minutes. Similarly,

maapvbbars

afwtl010.)

547-5271/$ -see front matter © 2010 Heart Rhythm Society. All rights reserved

nds). DF and CFAE sequential maps failed to identify 93.0% �2.4% and 35.9% � 14.9% of DF and CFAE sites, respectively.

ONCLUSION Because of temporal variability, sequential DF andFAE maps do not accurately reflect the spatial distribution ofxcitation frequency during any given sampling interval. The spa-ial distribution of DF and CFAE sites on maps created with se-uential point acquisition depends upon the time at which eachite is sampled.

EYWORDS Atrial Fibrillation; Fractionated electrograms; Fre-uency analysis; Mapping; Electrogram; Temporal stability; CFAE;ominant frequency

BBREVIATIONS AF � atrial fibrillation; AFCL � atrial fibrilla-ion cycle length; AIPI � average inter-potential interval;FAE � complex fractionated atrial electrograms; CL � cycleength; CV � coefficient of variation; DF � dominant frequency;F � high frequency; ICL � interval confidence level; SD � standardeviation

Heart Rhythm 2010;xx:xxx) © 2010 Heart Rhythm Society. All rights

FAEs were transient (average duration of CFAE 8.8 � 11.3 sec- reserved.

ntroductiont has been postulated that mapping of dominant frequencyDF)1 or complex fractionated atrial electrograms (CFAE)2

an identify target sites for ablation in patients with atrialbrillation (AF). In DF mapping, focal drivers are identifiedased upon radially distributed frequency gradients.1,3

FAE mapping identifies repetitive low-amplitude high-requency (HF) electrograms as ablation targets.2 In both

This study was supported by a research grant from Medtronic andducational grants from Boston Scientific and St. Jude Medical. PS con-ults for Biosense Webster and Medtronic. Address reprint requestsnd correspondence: Peter Spector, M.D., McClure 1 Cardiology, 111olchester Avenue, Burlington, Vermont 05401. E-mail address: peter.

pector@uvm.edu. (Received October 28, 2009; accepted January 6,

apping strategies, electrogram recordings are sequentiallycquired with a roving catheter and thus are valid only if DFnd CFAE remain stable over the timescale of the mappingrocedure. We postulated that frequency and fractionationary over time (Figure 1) and that, accordingly, the distri-ution of DF or CFAE on a sequentially acquired map wille influenced by the times at which points are sampled. Asconsequence, sequential mapping may yield misleading

esults regarding the loci involved in the initiation or per-istence of AF.

Recording from multiple sites simultaneously allowsnalysis of spatial frequency distribution without the con-ounding effects of variation in time. In the present study,e acquired spatially stable multielectrode recordings with

he use of a 64-electrode basket catheter. By virtue of

eaving electrodes in the same positions for 5 minutes, we

. doi:10.1016/j.hrthm.2010.01.010

wtdDtaC

MSEdstvpR

EAsdafrAptMmm

atmtbbstPc

SWrbpr

DBta(na

SDDddp5aw

FFdbenafq

Fs more f

T

AFCHDSAA

L

2 Heart Rhythm, Vol xx, No x, Month 2010ARTICLE IN PRESS

ere able to determine the stability of frequency and frac-ionation over much longer time intervals than with a stan-ard approach to mapping (�5 seconds/site). We createdF and CFAE maps using either simultaneously or sequen-

ially acquired data. From these maps, we quantified theccuracy of sequential mapping for identification of DF andFAE sites.

ethodstudy populationighteen patients presenting for ablation of symptomaticrug-refractory AF were studied (six paroxysmal, 10 per-istent, two long-standing persistent). The characteristics ofhe patients are summarized in Table 1. All patients pro-ided written and informed consent. The study was ap-roved by the Institutional Review Committee on Humanesearch.

lectrophysiologic studiesll patients were anticoagulated for �1 month before the

tudy. Antiarrhythmic medications were stopped at least 5ays before the study, and amiodarone was discontinued fort least 3 weeks. Patients were anesthetized with desflurane,entanyl, and rocuronium and intubated. Patients were en-olled in the study only if they had spontaneous or inducedF that lasted for at least 5 minutes. A double transseptalrocedure was performed, and a 64-electrode basket cathe-er (1.25-mm ring electrodes; Boston Scientific, Natick,

A) was placed in the left atrium and positioned to maxi-ize the number of electrodes in stable contact with theyocardium (Figure 2). Basket size was chosen based on

igure 1 Example of the temporal variability of frequency and fractiolower and less fractionated in the beginning of the tracing and faster and

able 1 Patient characteristics (n � 18)

ge, mean � SD 59.7 � 9.7emale sex (%) 6 (33)oronary artery disease 1 (5)ypertension 11 (59)iabetes mellitus 2 (9)leep apnea 2 (9)F duration, months 61.5 � 73.4F type:Paroxysmal 6 (33)Persistent 10 (55)Long-standing persistent 2 (11)

eft ventricular ejection fraction 58.1 � 12.8

gNote: all data are mean � SD or n (%).

trial dimensions as measured on a preprocedural computedomography scan. Basket dimensions were 38, 48, 60, or 75m with interelectrode spacing of 3, 4, 5, or 7 mm, respec-

ively. Initial catheter position was recorded with the use ofiplane cineangiography and then periodically checked withiplane fluoroscopy to confirm stability. In five randomlyelected patients, electrode positions were monitored con-inuously with the NavEx system (St. Jude Medical, St.aul, MN). Electrode movement was analyzed off-line andompared with the temporal changes in DF and CFAE.

tudy protocolith the catheter in a stable position, electrograms were

ecorded for 5 minutes. After acquisition of the data, theasket catheter was removed. The planned clinical ablationrocedure was then performed without regard to the basketecordings.

ata acquisitionipolar signals were sampled at 1 kHz and filtered from 30

o 500 Hz (Bard EP, Lowell, MA). Signals were exportednd analyzed off-line with the use of MATLAB softwareMathWorks, Natick, MA). Channels with obviously tech-ically compromised signals were excluded from furthernalysis.

ignal processing and spectral analysisF analysisF analysis was performed with the use of the methodsescribed by Sanders et al.1 Signals were Hanning win-owed, rectified, and band-pass filtered (3–15 Hz). A 4096-oint fast Fourier transformation was performed for each-second interval. The DF was defined as the maximummplitude in the magnitude squared spectrum. DF valuesere excluded if the regularity index was less than 0.2.

requency mapsor each patient, we used the DF values calculated asescribed above to generate two kinds of frequency distri-ution maps: simultaneous maps in which the frequency atach site (bipolar electrode pair) was calculated from sig-als acquired during the same 5-second recording period forll 32 electrode pairs and sequential maps in which therequency at each electrode pair was calculated from se-uential 5-second recording periods. A random number

ecorded on a bipolar electrogram from the basket catheter. Activation isractionated in the middle and slower by the end.

nation r

enerator was used to select the order in which points were

cFaSvn

DUDcta

CEfai0oswsi

CAop

FTobttmpas

aapb

EcDoeuDpuwspI(o

AFtssD1woafi

SDacnm

e cinea

3Habel et al Temporal Variability of Frequency and FractionationARTICLE IN PRESS

hosen for the sequential maps (location and time segment).or each patient, 10 separate sequential maps were gener-ted with different sequences of data points for each map.imultaneous maps were produced for each 5-second inter-al during the 5-minute recording period (i.e., 60 simulta-eous maps per patient).

F sitessing the methods of Sanders et al,1 a site was defined as aF site if its DF was at least 20% greater than the DF of its

losest neighbors (i.e., the eight electrode pairs adjacent tohe recording site). Sites were included only if data werevailable from three or more of the nearest neighbor sites.

FAE analysislectrograms were analyzed with custom software modified

rom a commercially available automated CFAE detectionlgorithm (Biosense Webster, Diamond Bar, CA) and dividednto 2.5-second segments. Deflections that were greater than.05 mV but less than 0.15 mV were tagged.4,5 The durationf the intervals between these deflections was then mea-ured. Intervals greater than 70 ms but less than 120 msere defined as CFAEs.5 The number of CFAEs per 2.5-

econd segment (interval confidence level [ICL]) and thenterpotential intervals were recorded.

FAE sitessite was defined as a CFAE site based upon two previ-

usly reported definitions: ICL �55 and an average inter-otential interval (AIPI) of �100 ms.6

alse-positive and false-negative siteshe purpose of DF and CFAE mapping is the identificationf sites that participate in maintaining AF so that they cane targeted for ablation. As such, any site that demonstrateshe potential to drive AF at any time qualifies as an ablationarget. For our analysis, we therefore considered sites thatet the above criteria for DF and CFAE sites as “true

ositives” if they were positive on the simultaneous map atny time throughout the 5-minute recording. Accordingly, a

Figure 2 Left anterior oblique and right anterior obliqu

ite was defined as false negative if it was not identified as p

DF/CFAE site on the sequential map but was identified onny of the simultaneous maps. A site was considered a falseositive if it was identified as a DF site on a sequential maput not on any of the simultaneous maps.

lectrode movement compared with temporalhanges in DF and CFAEF, AIPI, and ICL were measured using a sliding window,ffset by 13 ms. The x, y, and z coordinates for eachlectrode (sampled every 13 ms by the NavEx system) weresed to calculate its position relative to its initial location.F and CFAE measurements as well as position data werelotted against time and then underwent frequency analysissing fast Fourier transformation. The frequency resolutionas adapted to the sampling frequency and the number of

ampled data points. Linear correlation analysis was thenerformed on the resulting power spectra (DF, AIPI, andCL vs. electrode position). Mean and standard deviationSD) of Pearson’s correlation coefficient were calculatedver all sites.

nalyses for comparison with previous studiesor the purpose of direct comparison with prior studies of

he temporal stability of DF and CFAE, we performedeveral analyses analogous to those performed in thesetudies. For comparison with Sanders et al,1 we examinedF during the first and second 5-second time segments of0 randomly selected 10-second intervals. For comparisonith Scherr et al,6 we analyzed 10 randomly selected pairsf time points from each map location in all 18 patients. Inddition, for comparison with Verma et al,7 we analyzed therst time sample from each minute of our data.

tatistical analysisata are reported using descriptive statistics (mean � SD),

nd to further describe the dispersion of measurements, theoefficient of variation (CV) was calculated. The CV is aormalized measure defined as the ratio of the SD to theean. Wilcoxon signed-rank test and Friedman test were

ngiogram of the basket catheter placed in the left atrium.

erformed with the PRISM (version 5.01) statistical soft-

wc

RRWts2

DWsitfF

TTcctari(F

STftDveb

DSs5wo

tnlspt

DAsT(asfat

ir

FsZ

FCi

4 Heart Rhythm, Vol xx, No x, Month 2010ARTICLE IN PRESS

are package (GraphPad, Inc., La Jolla, CA). P�.05 wasonsidered statistically significant.

esultsecordingse recorded 32 bipolar electrograms per patient (64 elec-

rodes). DF and CFAE analysis was performed only onignals with adequate signal-to-noise ratio (an average of1.6 � 4.3 bipolar electrograms per patient).

Fe assessed the DF at each recording site during the 60

equential 5-second time segments of the 5-minute record-ng. Analysis of these DFs revealed marked variation fromime segment to time segment (temporal variability) androm recording site to recording site (spatial variability;igure 3).

emporal variability of DFo quantify temporal variability, we calculated the coeffi-ient of variation (SD/mean) over 5 minutes for each re-ording site. There was significant variation of DF overime: the average temporal coefficient of variation (all sites,ll patients) was 22.7% � 5.4%. As indicated by the SD andange of the coefficients themselves, there were also signif-cant differences in the amount of variation from site to siteP�.02 for all patients) and patient to patient (P � .008;igure 4).

patial variability of DFhe spatial coefficient of variation across all recording sites

or each 5-second segment was calculated to quantify spa-ial variability. This analysis revealed marked variation ofF across the atrium: the average spatial coefficient ofariation was 26.9% � 6.6%. There were significant differ-nces between time segments (P�.0001 for all patients) andetween patients (P � .008).

igure 3 Example of DF variability (in patient 15) plotted over time andpace. Graphed from right to left on each spline are electrode pairs 1 to 4.

beros indicate poor signal quality or regularity index �0.2.

F site identification was time dependentimultaneous maps were created of each 5-second timeegment (i.e., 60 maps were created for each patient; Figure). As a consequence of temporal variation of DF, DF sitesere present during some time segments but not duringthers.

An average of 11.4 out of 32 sites (36%) were iden-ified as DF sites after 5 minutes of continuous simulta-eous recording. DF sites were transient, on averageasting only 22.1 seconds out of 5 minutes. Therefore,election of a single 5-second sample by sequential map-ing would yield only a 7.4% probability of identifyinghese DF sites.

F site identification was sequence dependents a result of temporal variability, the value of DF at any

ite depended on the time at which that site was sampled.herefore, sequential maps could yield misleading results

e.g., by sampling fast areas when they are slow and slowreas when they are fast). Sequential maps differed fromimultaneous maps, sometimes correctly and sometimesalsely identifying DF sites. The 10 random sequential mapslso differed from each other, that is, the DF sites were nothe same on all of these 10 maps.

The probability that sequential mapping would fail todentify DF sites was 93%. In contrast, sequential mappingarely mistakenly labeled non-DF loci as DF sites (proba-

igure 4 Temporal CV of DF. The center square (�) indicates meanV over all sites; the thick gray bar indicates SD; and the thin black line

ndicates range.

ility 0.3%).

CWi2a

COCwcsnfsSo

CAawcocwi

TrTme

fvvtaw(

CCc

Ff(tc multane

Fott

5Habel et al Temporal Variability of Frequency and FractionationARTICLE IN PRESS

FAE vary over timee assessed the number of CFAEs and the interpotential

ntervals at each recording site during the 120 sequential.5-second time segments of our 5-minute recording. Bothnalyses revealed spatial and temporal variability.

FAE sites defined by ICL >5ut of all sites, 51% � 33% were identified as CFAE sites.FAE sites were transient; the average time to first CFAEas 11.1 � 37.1 seconds. The average duration that a site

ontinuously met criteria as a CFAE site was 8.8 � 11.3econds. The average duration for which a CFAE site didot meet criteria (i.e., the time during which it could bealsely classified as a “non-CFAE site”) was 6.8 � 15.7econds. CFAE site identification was sequence dependent.equential maps failed to correctly identify 35.9% � 14.9%f CFAE sites.

FAE sites defined by AIPI <100 mstotal of 47% � 41% of sites were CFAE sites. The

verage observation time required to first meet these criteriaas 10.8 � 36.6 seconds. The average duration that sites

ontinuously met CFAE site criteria was 47.2 � 79.7 sec-nds. The average duration that a CFAE site failed to meetriteria was 20.7 � 37.6 seconds. CFAE site identificationas sequence dependent; sequential maps failed to correctly

dentify 38% � 16% of CFAE sites.

emporal variations of DF and CFAE were notelated to catheter movementhe average electrode movement was �1.56, 0.98, and 1.73m along the x, y, and z axes, respectively. Analysis of

igure 5 Simultaneous versus sequential maps. (a) Example of a simulrom each electrode pair simultaneously acquired (blanks represent poor sitop to bottom � electrode pairs 1–4, left to right � spline a–h; note splineime segments each site). Top panel: DF values acquired sequentially. Boorrespond with those acquired simultaneously.) The DF site at C2 on the si

lectrode position versus time revealed a periodicity at the s

requency of respiration (all patients were mechanicallyentilated). In contrast, analysis of DF, AIPI, and ICLersus time revealed no periodicity. There was no correla-ion between electrode movement and changes in DF, AIPI,nd ICL, r � 0.24 � 0.20, 0.21 � 0.19, and 0.24 � 0.23ith P�.05 in 98%,100%, and 99% of all sites, respectively

Figure 6).

omparison with previous studiesonsistent with the findings of Sanders et al,1 when weompared only the first and second 5-second time segments

acquisition (same 5-second time segment all sites Top panel: DF valuesality or regularity index �0.2). Bottom panel: color-coded frequency mapcent to spline h). (b) Example of sequential acquisition (different 5-secondnel: color-coded frequency map. (Note that DF values do not necessarilyous map (*) is not identified on the sequential map (†; i.e., false negative).

igure 6 Examples of the frequency content of electrode movement andf change of DF over time. The red line represents the power spectrum ofhe DF versus time plot. The black line represents the power spectrum ofhe electrode position versus time plot. Linear correlation analysis reveals

taneousgnal qua is adjattom pa

ignificant noncorrelation (r � 0.2339, P�.001).

osrgdpasdfe1

DGhaofyotptciqsa

rtqcetdrinssbamaelati

tbset

ihDtmbdwfrCptpHtr

tsdppepsdtm

CSiWessoDat

llmasTAd

iaspt

6 Heart Rhythm, Vol xx, No x, Month 2010ARTICLE IN PRESS

f 10 randomly selected 10-second windows we found noignificant variability (5.55 � 1.82 vs. 5.55 � 1.84 Hz,espectively; P � .60). Similarly, in an examination analo-ous to that of Scherr et al,6 when we compared 10 ran-omly selected pairs of time points from each map (in allatients), we found that 83% � 3% of sites initially defineds CFAE sites (defined by AIPI �100 ms) remained CFAEites, while 17% � 3% did not meet criteria as CFAE sitesuring the second time point. Finally, similar to the resultsrom Verma et al,7 analysis of the first time sample fromach minute of our data revealed a mean �CL � 14.6 �4.1 ms (P � .10).

iscussioniven the prevalence of AF,8 its impact on patients and theealth care system,9 and the relatively poor efficacy ofntiarrhythmic medications,10,11 ablation has been vigor-usly pursued as a therapeutic approach. Indeed, ablationor AF has met with significant success over the past 10ears. Nevertheless, cure rates remain less than ideal,10,12

wing in part to our limited ability to accurately delineatehe spread of electrical activity during AF in each individualatient. Typically, ablation procedures begin with diagnos-ic mapping maneuvers designed to define the electricalircuitry underlying the arrhythmia. However, this approachs not used in ablation for AF because the activation se-uence during fibrillation is not temporally stable and theubstrate is functional and dynamic rather than anatomicnd fixed.

We believe that an ideal strategy for ablation of AFequires knowledge of its mechanism on a patient-by-pa-ient basis: in short, mapping for electrical guidance. Ade-uate mapping requires electrogram sampling with suffi-ient spatial and temporal resolution to identify variations inxcitation state. Unfortunately, because of practical limita-ions that constrain the number of electrodes that can beeployed transvenously, it is difficult to simultaneouslyecord from a sufficiently large number of atrial sites todentify activation patterns occurring during AF. In mostoncomplex arrhythmias, the challenge of adequate spatialampling is overcome by sequential mapping (so that aingle roving catheter can be used to sample a wide distri-ution of sites). The changing activation sequences associ-ted with AF, however, preclude valid sequential activationapping. Consequently, alternatives to the use of local

ctivation time have been sought for AF mapping. Twoxamples are DF and CFAE. The tacit presumption under-ying these approaches to AF mapping is that DF and CFAEre temporally stable and can therefore be sampled sequen-ially yet still accurately reflect underlying electrical activ-ty.

Using 5-minute recordings from 64 spatially stable elec-rodes, we were able to directly measure the temporal sta-ility of DF and CFAE over a wide distribution of left atrialites. In contrast, most previous studies of temporal stabilitymployed much shorter time windows and/or sampled in-

ermittently.1,6,7 We found that both DF and CFAE exhib- t

ted spatial and temporal variability. These observationsave implications with regard to the validity of sequentialF or CFAE mapping for identification of AF ablation

arget sites. As described by Haissaguerre and Jalife, DFapping identifies target sites based upon the relationship

etween activation frequencies at adjacent locations.1 Ourata demonstrate, however, that because frequency variesith time it is possible for a sequential map to display a

requency gradient that did not exist during any contempo-aneous interval (e.g., Figure 4). In the standard approach toFAE mapping, target sites are defined by the electrogramroperties recorded from a single location without regard tohe properties of adjacent sites. Hence, there can be no falseositives that result from the sequence of point sampling.owever, temporal variation can result in a failure to iden-

ify CFAE sites if points are sampled during a period ofeduced CFAE number or rate.

Most mapping systems acquire electrogram data sequen-ially, sampling approximately 5 seconds of signal at eachite. By contrast, we recorded a 60-fold greater duration ofata (5 minutes) at each site. Of course, the appropriateeriod required for adequate sampling is a function of theeriodicity of temporal variation, so we cannot be sure thatven 5 minutes per site is long enough to capture all tem-oral variation. Nevertheless, we found that new CFAEites became evident during this period of time. Thus, ourata acquired with the basket catheter demonstrate the po-ential of sequential mapping to generate inaccurate andisleading results.

omparison with previous studieschuessler et al13 made high-density multielectrode record-

ngs during AF in patients at the time of thoracotomy.13

ith the use of a 250-electrode array sutured to the atrialpicardium, they observed variation of DF during 2-secondegments of recording. Ninety-one percent of the recordingites exhibited a distinct region of maximum DF “at leastnce [emphasis added] during the recording period.” TheF site changed location in 48% of the patients studied. The

uthors noted that “the most common pattern observed inhe patients in this study was an unstable pattern.”

Rostock et al14 studied the relationship between AF cycleength (AFCL) and CFAEmax (defined as CFAE with theongest duration). They found that AFCL decreased by �10s before the occurrence of CFAEmax at 91% of left atrial sites

nd returned to baseline after CFAEmax.14 The basis of thistudy was the temporal variability of AFCL and CFAEmax.he fact that CFAEmax occurrence followed a decrease inFCL suggests that CFAEmax might reflect driven rather thanriver behavior.

Krummen et al3 used a basket catheter to examine driversn human AF. Their analysis demonstrated temporal vari-bility of DF. Simultaneous maps identified driver sites inix of 12 paroxysmal AF and five of 12 persistent AFatients. Three patients had two driver sites “at differentime intervals.” In two patients AF terminated during abla-

ion at driver sites identified by simultaneous maps. Of 1000

sd

DwdseTdsda.sd

�spp(ta1tbdrapro

wot(adwa

poiptc

CSordmd

hibisitisuOcs

LTnabTvctceaiwsipgpmawwfifvc

Dtota

CDtaraC

7Habel et al Temporal Variability of Frequency and FractionationARTICLE IN PRESS

imulated sequential maps, only 5.8% identified these sameriver sites.

Other investigators have studied the temporal stability ofF1 and CFAE 6,15–18 and concluded that they did not varyith time. We attempted to determine why our findingsiffered from these studies. For example, Sanders et al1

tudied DF in patients with AF presenting for ablation. Theyxamined temporal stability in a subset of these patients.hey found no temporal variability when comparing DFuring the first (6.2 � 1.8 Hz) and second (6.3 � 1.9 Hz) 5econds of 10-second recordings. When we examined ourata using the same analysis, we found no significant vari-bility (5.55 � 1.82 vs. 5.55 � 1.84 Hz, respectively; P �60). Thus, analysis of the full 5-minute data set demon-trates that brief periods of comparison are inadequate toetect temporal variability.

Scherr et al6 examined automated CFAE detection (AIPI100 ms) in 12 patients with persistent AF. To assess

hort-term CFAE stability, they acquired 2.5-second sam-les sequentially over 1 minute at 30 CFAE sites in sixatients. They found that 15.4 � 4.4 out of 17.5 � 3.587.9%) recordings were categorized as CFAEs. To inves-igate long-term stability, they revisited 387 of these sitesfter 34.3 � 8.7 minutes. Of the 149 sites defined as CFAE,35 (90.6%) remained CFAE sites on the second map. Ofhe 238 non-CFAE sites revisited, 13 (5.5%) were found toe CFAE sites during the repeat map. In analysis of ourata, 83% � 3% of sites initially defined as CFAE sitesemained CFAE sites while 17% � 3% did not meet criterias CFAE sites during the second time point. Again, a com-arison of only 5 seconds of data (two 2.5-second samples)eveals less variability than is apparent when all 5 minutesf recording are analyzed (i.e., undersampling error).

Verma et al7 examined CFAE stability in 24 patientsith persistent AF. In 12 patients, average CL was samplednce per minute for 5 minutes. For sites with a mean CL inhe 50–120 ms range, there was no significant changemean �CL � 12 � 4 ms, P � .4). Once again, in our data,nalysis of the first time sample from each minute of ourata revealed a mean �CL � 14.6 � 14.1 ms (P � .10). Itas only during analysis of 5 minutes that variability was

ppreciated.Although our data indicate statistically significant tem-

oral variability of local AF frequency, it is the comparisonf sequential and simultaneous maps that revealed the clin-cal significance of temporal variation. These data reveal theotential for the targeting of inappropriate sites and failureo target appropriate sites when CFAE or DF maps arereated by sequential acquisition.

linical implicationsuccessful mapping of AF will likely improve the efficacyf ablation but remains elusive. The development of accu-ate mapping techniques is likely to be facilitated by clearlyefining current impediments. The challenges to optimalapping result from the complexity of AF itself. The spatial

ensity of changes in activation during AF requires a very a

igh concentration of recording sites. This, combined withncreased atrial size, mandates the use of a very large num-er of recording electrodes. In addition, the temporal variabil-ty of activation during AF precludes the use of sequentialampling to create activations maps. Because of practical lim-tations on the number of electrodes and amplifier channelshat can be used, adequate multisite simultaneous recordings faced with potentially insurmountable obstacles. In lieu ofuch mapping techniques, we remain largely dependentpon an anatomic approach to ablation for therapy of AF.ur results demonstrate the potential pitfalls that may en-

umber sequential mapping of DF or CFAE for targeting ofites to ablate during AF.

imitations in our studyo assess the temporal stability of CFAE and DF, weeeded to obtain simultaneous recordings from multipletrial sites. Accordingly, we used a commercially availableasket catheter that allowed recording from 64 electrodes.his catheter is well suited for identification of temporalariability for the purposes of characterizing potential inac-uracies encountered with sequential mapping. However,his catheter is ill suited as a practical clinical mappingatheter for several reasons. Given its geometric design,ndocardial contact can be difficult to achieve in some lefttrial locations, especially the floor and septum. In addition,t is frequently not possible to make endocardial contactith all of the electrodes. Furthermore, the interelectrode

pacing and spatial distribution of electrodes is less thandeal. However, these well recognized limitations do notreclude our ability to address the question under investi-ation—namely, whether electrogram frequency (and com-lexity) varies over time during AF and whether sequentialapping can yield misleading results. It is possible that DF

nd CFAE exhibit greater temporal stability at sites thatere not sampled in this study (e.g., the left atrium floor orithin the pulmonary veins). However, because we identi-ed temporal variability at sites that are frequently targetedor ablation (left atrial roof, posterior wall, and pulmonaryein antrum), these data provide important information forlinical electrophysiologists.

Our data do not address the electrophysiologic question:oes DF or CFAE mapping accurately identify sites critical

o the perpetuation of AF? Rather our data indicate that ifne does use DF or CFAE mapping, then acquisition ofhese maps via sequential sampling leads to significant in-ccuracies.

onclusionsF and CFAE mapping have been proposed as alternatives

o activation mapping. However, because of temporal vari-bility, sequential DF and CFAE maps do not accuratelyeflect the spatial distribution of excitation frequency duringny given sampling interval. As a result, sequential DF orFAE mapping can result in failure to target or inappropri-

te targeting for ablation.

R 1

1

1

1

1

1

1

1

1

8 Heart Rhythm, Vol xx, No x, Month 2010ARTICLE IN PRESS

eferences1. Sanders P, Berenfeld O, Hocini M, et al. Spectral analysis identifies sites of

high-frequency activity maintaining atrial fibrillation in humans. Circulation2005;112:789–797.

2. Nademanee K, McKenzie J, Kosar E, et al. A new approach for catheter ablationof atrial fibrillation: mapping of the electrophysiologic substrate. J Am CollCardiol 2004;43:2044–2053.

3. Krummen DE, Peng KA, Bullinga JR, Narayan SM. Centrifugal gradients of rateand organization in human atrial fibrillation. Pacing Clin Electrophysiol 2009;32:1366–1378.

4. Wu J, Estner H, Luik A, et al. Automatic 3D mapping of complex fractionatedatrial electrograms (CFAE) in patients with paroxysmal and persistent atrialfibrillation. J Cardiovasc Electrophysiol 2008;19:897–903.

5. Scherr D, Dalal D, Cheema A, et al. Automated detection and characterizationof complex fractionated atrial electrograms in human left atrium during atrialfibrillation. Heart Rhythm 2007;4:1013–1020.

6. Scherr D, Dalal D, Cheema A, et al. Long- and short-term temporal stability ofcomplex fractionated atrial electrograms in human left atrium during atrialfibrillation. J Cardiovasc Electrophysiol 2008;1:13–21.

7. Verma A, Wulffhart Z, Beardsall M, et al. Spatial and temporal stability ofcomplex fractionated electrograms in patients with persistent atrial fibrillationover longer time periods: relationship to local electrogram cycle length. HeartRhythm 2008;5:1127–1133.

8. Go AS, Hylek EM, Phillips KA, et al. Prevalence of diagnosed atrial fibrillationin adults: national implications for rhythm management and stroke prevention:the AnTicoagulation and Risk Factors in Atrial Fibrillation (ATRIA) Study.JAMA 2001;285:2370–2375.

9. Benjamin EJ, Wolf PA, D’Agostino RB, et al. Impact of atrial fibrillation onthe risk of death: the Framingham Heart Study. Circulation 1998;98:946 –

952.

0. Calkins H, Reynolds MR, Spector P, et al. Treatment of atrial fibrillationwith anti-arrhythmic drugs or radio frequency ablation: two systematicliterature reviews and meta-analyses. Circ Arrhythmia Electrophysiol 2009:CIRCEP.108.824789.

1. Nair GM, Nery PB, Diwakaramenon S, et al. A systematic review of randomizedtrials comparing radiofrequency ablation with antiarrhythmic medications inpatients with atrial fibrillation. J Cardiovasc Electrophysiol 2009;20:138–144.

2. Fuster V, Ryden LE, Cannom DS, et al. ACC/AHA/ESC 2006 guidelines for themanagement of patients with atrial fibrillation: full text: a report of the AmericanCollege of Cardiology/American Heart Association Task Force on practiceguidelines and the European Society of Cardiology Committee for PracticeGuidelines (Writing Committee to Revise the 2001 guidelines for the manage-ment of patients with atrial fibrillation) developed in collaboration with theEuropean Heart Rhythm Association and the Heart Rhythm Society. Europace2006;8:651–745.

3. Schuessler RB, Kay MW, Melby SJ, et al. Spatial and temporal stability of thedominant frequency of activation in human atrial fibrillation. J Electrocardiol2006;39:S7–12.

4. Rostock T, Rotter M, Sanders P, et al. High-density activation mapping offractionated electrograms in the atria of patients with paroxysmal atrial fibril-lation. Heart Rhythm 2006;3:27–34.

5. Takahashi Y, Sanders P, Jais P, et al. Organization of frequency spectra of atrialfibrillation: relevance to radiofrequency catheter ablation. J Cardiovasc Electro-physiol 2006;17:382–388.

6. Stiles MK, Brooks AG, Kuklik P, et al. High-density mapping of atrial fibril-lation in humans: relationship between high-frequency activation and electro-gram fractionation. J Cardiovasc Electrophysiol 2008;19:1245–1253.

7. Lin YJ, Tai CT, Kao T, et al. Consistency of complex fractionated atrialelectrograms during atrial fibrillation. Heart Rhythm 2008;5:406–412.

8. Roux JF, Gojraty S, Bala R, et al. Complex fractionated electrogram distribution

and temporal stability in patients undergoing atrial fibrillation ablation. J Car-diovasc Electrophysiol 2008;19:815–820.