Mid- and Long-Term Similarity of VentricularResponse to Paroxysmal Atrial Fibrillation:Digoxin versus Placebo

KATERINA HNATKOVA, FRANCIS D, MURGATROYD, JAN POLONIECKI,JOHAN E,P. WAKTARE, CLIF A. ALFERNESS,* A. JOHN CAMM,and MAREK MALIK

From the Department of Cardiological Sciences, St. George's Hospital Medical School. London, England;and *InControl Inc.. Redmond, Washington

HNATKOVA, K., ET AL.: Mid- and Long-Term Similarity of Ventricular Response to Paroxysmal AtrialFibrillation: Digoxin versus Placebo. The effects of digoxin on ventricular response during atrial fibrilla-tion (AF) and consequent effects on arrhythmic symptoms have stiU not been fully explained. This studyinvestigated whether the treatment by digoxin contributes to mid- and long-term stabilization of ventric-ular cycles in patients with paroxysmal AF. A population of 45 patients with paroxysmal AF underwent24'hour FCG recordings during each arm of a raadomizeid crossover trial comparing digoxin and placebo.This yielded 30 Holter recordings from 22 patients that contained AF episodes lasting in excess of 2 min-utes and with acceptably low Holter noise. Each AF episode was divided into nonoverlapping segments of30 seconds and the distribution ofHR intervals in each segment was compared with the distribution of allother AF segments in the same recording using the Kolmogorov-Smirnov test. The percentage of tests thatrevealed significant differences at levels ofP ^ 0.01, and P < 0.001 were sorted according to the time be-tween the segments compared. The comparisons of these results were performed between: (a) all record-ings on placebo (n = 16] and all recordings on digoxin (n = 14). and (b) between recordings on placeboand on digoxin in 8 patients in whom paired analysis was possible. Adjacent AF segments (distance 0)differed significantly only in < 30% of both recordings on placebo and on digoxin. However, with in-creasing tho distance between segments, the proportion of the significant differences between RU intervaldistributions increased more with placebo than with digoxin (P < 10'-'"". Chi-square test). Paired data re-vealed larger differences between placebo and digoxin with increasing distance between segments. Thusin patients with paroxysmal AF, digoxin leads to more reproducible patterns of ventricular cycles thatmaybe better tolerated clinically. (PACE 1998; 21:1735-1740}

long-term ECG, atrial fibrillation, digoxin, rhythm self-similarity, KoimagoroV'Smirnov test

Introduction

Digoxin is frequently prescribed to patientswith paroxysmal atrial fibrillation {AF) and is gen-orally helieved to reduce the symptomatic impactof the disorder. Several hypotheses have been pro-posed to justify this clinical practice. Most fre-quently, reduction of the ventricular rate during

Addre.ss for reprints; Katerina Hnatkova, M.D., Department ofCardioloHical Sciences, St. Gcorgo's Hospital Medicat School,Crannier Terrace, London SW17 ORE, England. Fax: 44-181-7()7-7141; e-mail: k.hnatkovaCysghms.ac.ukReceived November 12, 1996; revised May 5, 1997; acceptedSeptember 30, 1997,

paroxysms of AF has been considered to he themain mechanism of benefit. However, recent clin-ical studies have failed to support this conc:ept''^and the continued use of digoxin in paroxysmalAF is questioned.' Although the clinical practiceof prescribing digoxin to patients witb paroxysmalAF is not founded in scientific evidence, it is pos-sible that a benefit of digoxin may arise from amore subtle effect than the simple reduction ofventricular response during AF paroxysms.

With this in mind, this study investigatedwhether the treatment by digoxin contributes tomid- and long-term stability of ventricular cyclesin patients with paroxysmal AF.

PACE, Vol. 21 September 1998 1735

HNATKOVA, ET AL.

MethodsData and Recordings

The study population consisted of a subset ofpatients who had been recruited in a multicenterstudy investigating the effects of different modesof pharmacological treatment of paroxysmal AF(CRAFT 1 study)." All patients in this study un-derwent 24-hour ambulatory Holter monitoring(two-channel recordings of leads II and modifiedCM5) during each phase of a double-blind,crossover trial when digoxin and placebo weregiven in a random order. Digoxin dosing was ad-justed to achieve therapeutic plasma levels, afterwhich monitoring commenced. The mediandigoxin dose was 375 |j.g/day; the mean plasmalevel was 1.01 (± 0.27) fxg/L-

A database of 86 Holter recordings was col-lected from 45 patients (25 men, mean age at en-rollment of 59.3 ± 12.1 years) in whom paroxysmsof AF had been noted in at least one recording. Ofthese, 23 patients were recorded on placebo, and22 on digoxin. Recordings containing only AF oronly sinus rhythm episodes were excluded fromthe analysis, as were recordings which were of in-sufficient quality to allow useful analysis.

For the purposes of this investigation, onlyrecordings containing an AF episode of at least 2-minute duration and that were polluted by < 20%of noise or Holter artifact were included.

Episodes of AF were identified hy a previouslyvalidated semiautomated method.^ In brief, eachrecording was subjected to an analysis and manualediting using a commercial Holter system (LaserHolter System 8000, Marquette Medical SystemsInc., Milwaukee, WI, USA). The precise timing ofeach episode of AF and sinus rhythm was deter-mined hy visual inspection on a full small-scaleprintout of the total recording, and marked using adigitizing board. These data were matched with thecorresponding heats in a conventional RR intervalfile generated by Holter analysis. A composite filelisting the rhythm of each beat, RR interval dura-tion, and the marker of noise in every RR intervalwas created. In addition, the file listed the real-time of each beat, and the QRS complex morphol-ogy. The validation study of this method demon-strated that the onset and termination of each AFepisode can be identified within the Holter beat filewith a precision of ± 1 cardiac cycle.

Rhythm Analysis

Each Holter recording was divided into por-tions of paroxysmal AF and sinus rhythm. Eachepisode of AF of at least 2-minute duration was di-vided into nonoverlapping portions of 30 seconds.These portions are termed 'segments.' Holter arti-fact and recognition noise were evaltiated withineach segment. Segments containing more than20% (in duration) noise or artifact were excludedfrom subsequent analysis.

For each segment, the distribution of RR in-terval durations was constructed. This empiricaldistrihution was compared with that of all otherAF segments in the same recording (i.e., each seg-ment was compared to other segments not onlywithin the same AF episode but also to all othersegments from other AF episodes within the samerecording) using the Kolmogorov-Smirnov test. Inhrief, the empirical distribution function E of anobservation set [xi, X2, . . . , x ] that has been or-dered (xi < Xi + i for 1 < i < n) is defined as

Ex(a) = 0 for a < Xi,

Ex(a) = k/n for x^ a < xj +i,

1 < k < n ,

Ex(a) = 1 for a > XR.

The Kolmogorov-Smirnov test compares twoempirical distribution functions. The statistics ofthe test is based on the maximum vertical distancebetween both distrihution functions. This test ismore powerful than the t-test that is based on com-parisons of mean values of two sets of observa-tions. It is also more powerlnil than nonpararaetrictests that compare two sets of observations basedon the rank of all values and describe how sys-tematically are values of one set lower than valuesof the other. Figure 1 demonstrates the perfor-mance of the Kolmogorov-Smirnov test and showsan example with real data from this study inwhich the Kolmogorov-Smirnov test (hut not the /-test or the Mann-Whitney test) was able to docu-ment the differences between the patterns of thesegments compared.

Considering two different levels of statisticalsignificance (P < 0.01 and P < 0.001). the indi-vidual Kolmogorov-Smirnov tests were classified

1736 September 1998 PACE, Vol. 21

HEART RATE PATTERNS DURING ATRIAL FIBRILLATION

RR Interval Duration [ms]1,000

800

600

400

mean RR - 431.9 msSD RR - 96.9 ms

A14 28 42 56

RB Interval Position [#RR]

70

RR Interval Duration [ms]1.000

800

600

400

200

mean RR = 431.9 msSD RR = 97.5 ms

s • • - •

B14 28 42 56

RR Interval Position [#RR]

70

Cummulative Probability100 %

80 % -

60 %

40 %

20 %

0 %

Hest: p = 1.0MW-test: p - 0.48KS-test: p < 0.001

300 450 600

RR interval [ms]

750

Figure 1. Example of the analysis in a record taken in aSi-year-old woman. Panels A and B show examples ofRR tachograms in atrial fibrillation (AF) segments thatshow different patterns of similarity. The vahies of mean

into those revealing and those not revealing a sig-nificant difffirence between compared distribu-tions. The time shift, rounded to minutes, hetweenthe beginnings of each pair of compared segmentswas taken as the distance hetween segments. Thedistance of adjacent segments within the sameepisode was 0 minutes.

For each distance, the proportion of segmentswith significantly different RR interval distribu-tions was calculated between all segments com-pared. For each recording, this resulted in a graphgiving the percentage of statistically significanttests (at P < 0.01 and P < 0.001. respectively) foreach distance between the segments compared. Indifferent groups of recordings [see further) thesegraphs were averaged and represented by hands ofmean ± standard error. Global graphical character-istics of different groups were ohtained in this way.

For each distance, the numbers of signifi-cantly different tests on digoxin versus on placebowere compared using the standard Chi-square test.This comparison was performed twice: (a) be-tween all recordings on placebo and all recordingson digoxin (pooled data), and (b) between record-ings on placebo and on digoxin in those patientsfor whom both the recording on placebo and ondigoxin were eligible for this study (case con-trolled data).

Similar analyses were performed using stan-dard tests evaluating differences in mean and stan-dard deviation of RR interval duration of individ-ual segments. The difference between twosegments was evaluated using: (a) /-test for the com-parisons of mean RR intervals, and (b) F-test for thecomparisons of standard deviations of RR intervals.

In practical terms, the proportion of tests withstatistically significant differences between the

and standard deviation of RR interval durations arenearly the same (431.9 ms and 96.9 ms in pane] A. and431.9 ms and 97.5 ms in panel B). Standard \-test as wellas the more sensitive nonparametric Mann-Whitney test(MW) did not find any significant differences betweenboth series of HH intervals. Kolmogorov-Smirnovstatistics (panel C) performed differently. By evaluatingthe maximum distance between both distributions, itclassified them as significantly different (P < 0.001).which is more relevant.

PACE, Vol. 21 September 1998 1737

HNATKOVA, ET AL.

comparod distributions represent a measure of sta-bility of the pattern of ventricular response duringparoxysms of AF (i.e., the more stable rhythm, thefewer the tests showing significant differences he-tween distributions in compared segments). Thedistance between the segments gives a measure ofthe short-, mid-, and long-term stahility.

Results

Of all the Holter tapes available in thedatabase, 30 recordings (14 on digoxin and 16 on

placebo) made in 22 patients (age 59.2 ± 13.3years, 12 men) satisfied the selection criteria. Ofthese, paired recordings (i.e.. both on digoxin andplacebo) were available in eight patients (aged55.5 ± 9.9 years, 5 men). In total, the study inves-tigated 130 AF episodes with 12,619 segments; ofthese, 86 episodes (6,993 segments) were recordedon placebo and 44 episodes (5.626 segments) ondigoxin. The mean duration of all qualifyingepisodes was 52.2 ± 126.9 minutes (median 9.2mins). The mean duration of qualifying episodesrecorded on placebo and on digoxin was 81.7 ±

Proportion of segments different at p=0,01100

120 240 360 460 600

Distance of segments [min]

720

Proportion of segments diffe-ent at p = 0.01120

120 240 360 480 600

Distance of segments [min]

720

80-

60

40

20

Proportion

k

of segments

1M

different

r%

at p

J

= 0-001 Proportion of segments different at p = 0.001

B

120 240 360 480 600Distance of segments [min]

720

120

D

120 240 360 480

Distance of segments [minj

600 720

Figure 2. Shows the percentages of atrinl fibrillation segments with RR inten'nl distributionsdifferent at the level of P ^ 0.01 (panels A and Cj. and ^ 0.001 (panels B and D). The full bandscorrespond to recordings on placebo, open bands to digoxin (mean ± SE). The linear interpolationof trend of the data is shown in full fine line for digoxin. and in dashed line for placebo. PanelsA and B show the results for pooled data of all recordings on placebo and digoxin. Panels C andD show the results for case controlled data (see the text for details).

1738 September 1998 PACE, Vol. 21

HEART RATE PATTERNS DURING ATRIAL FIBRILLATION

175.5 minutes and 36.5 ± 88.2 minutes (P = NS.nonparametric Wilcoxon test), respectively.

The mean value of RR intervals during quali-fying segments was 550.9 ± 140.0 ms in patientswho were recorded on digoxin, and 524.6 ± 134.3ms in those recorded on placebo (P < 10" *, non-parametric Wilcoxon test). The standard deviationof RR intervals during qualifying segments was129.4 ± 49.1 ms in patients recorded on digoxin,and 115.6 ± 45.2 ms in those on placebo (P <lO"* , nonparametric Wilcoxon test). Thus, in ourdata, the global distributions of RR interval dura-tions on placebo and digoxin are principally dif-ferent. Unfortunately, these comparisons per-formed on pooled data of individual segments donot offer a representative statement about ventric-ular rate on placebo and digoxin because of tbedifferent contribution to pooled data and bias in-troduced by different number of segments in indi-vidual recordings.

Figure 2 sbows tbe percentages of AF seg-ments with RR interval distributions different atthe levels of P < 0.01 and P < 0.001. Panels A andB show the global characteristics of pooled datawith distributions different at significance levelsof P < 0.01 and P < 0.001, respectively. Similarly,panels C and D present the global characteristic ofcase controlled data at the significance levels of P< 0.01, and P < 0.001, respectively.

The adjacent segments of AF rhythm differedsignificantly (P :^ 0.01) in < 35% cases, both forthe segments belonging to the recordings onplacebo and on digoxin. With increasing the levelof statistical difference to P < 0.001. the propor-tion of different adjacent segments was reducedbelow 25%. However, with increasing distancebetween segments, tbe proportion of significantdifferences between RR interval distributions in-creased more with placebo tban witb digoxin inbotb pooled and case controlled data. After theinitial 100 minutes, curves corresponding toplacebo and digoxin were clearly separated. How-ever, the differences between digoxin and placeborecordings were stronger with increasing distancebetween segments in case controlled data. Theanalysis of case controlled data revealed oppositerelation (i.e., higher proportion of different seg-ments for digoxin than for placebo) between 55and 90 minutes of segment distance. Thus,digoxin may bave a substantial influence on rate

stability in the mid- and long-term, but not neces-sarily in the short-term. Tbe Chi-square test re-vealed a level of statistical significance of up to P< 10 •'"" between both curves for both pooled andcase controlled data.

Tbe comparisons of individual segmentsbased on simple mean and SD RR were performedusing the standard /-test and F-test, and both testsrevealed similar results for pooled and case con-trolled data. Although the trend of curves ofplacebo and of digoxin was similar to tbat of thecharacteristics obtained with Kolmogorov-Smirnov statistics, comparisons hased on Kol-mogorov-Smirnov test were more sensitive.

Discussion

Digoxin is a drug commonly prescribed toimprove the symptoms of paroxysmal AF, pre-dominantly with the aim of controlling ventricu-lar rate during AF episodes.** However, most stud-ies have failed to demonstrate a beneficial effect.Rawles et al.^ retrospectively studied 72 patientswith paroxysmal AF, and found that the ventricu-lar rate at the onset of an episode of paroxysmalAF was no lower if the patient was on digoxin.This remained true for tbe duration of tbeepisode, and the standard deviation of the heartrate did not differ hetween patients on digoxinand others. In addition, episodes of paroxysmalAF that occurred on digoxin were four times aslikely to last more tban 30 minutes (P < 0.05).Galun et al.^ found no difference in the meanheart rate of AF on admission to hospital in pa-tients with a previous history of AF of recent on-set between those taking digoxin and those not.However, both of tbese were observational stud-ies without randomized treatment assignment. InAF of recent onset, Falket al.^ found no differencebetween digoxin and placebo in a randomized,double-blind trial in terms of cardioversion rates.

Based on published evidence, the continuedpopularity of digoxin is therefore eitber erro-neous, or results from subtle beneficial effectstbat are not detectable by simple statistical mea-sures. Our observations provide support for thesecond possibility. Improved medium to long-term (over 100 mins) stability may lead to morestable myocardial function as a result of less vari-ability of ventricular performance. Tbe ohserva-

PACE.V0L21 September 1998 1739

HNATKOVA, ET AL.

tions from this study may also be interpreted assuggesting that digoxin increases the self-similar-ity rather than slowing down the ventricular re-sponse.

Indeed, the fact that we observed stronger dif-ferences using Kolmogorov-Smirnov test thanwith f-test and F-test may he hased not only on thegeneral strength of this test. Perhaps, by increasingthe self-similarity of the rhythm, digoxin affectsmore the general pattern of the ventricular re-sponse during AF rather than merely the meanventricular rate or its global variability.

A self-similar rhythm would he expected toimprove both symptoms at rest and to blunt thesudden change in rhythm in response to physio-

logical stress. Intuitively, short-term self-similar-ity would he expected to be most important, butour findings actually pointed to more effect in themedium- to long-term. In order to prove the rele-vance of this finding, the ohscrved drug effectwould need to be correlated with clinical benefit(e.g., exercise capacity), and the magnitude oftheobserved effect compared with other agents usedto treat paroxysmal AF.

Hence, the study seems to support the hy-pothesis that digoxin causes subtle long-term self-similarity of ventricular periods in patients withparoxysmal AF. It is possible to speculate that thismay cause AF paroxysm to be clinically better tol-erated.

References

1. Rawles JM, Metcalfe M], Jennings K. Time of occur-rfince, duration, and ventricular rate of paroxysmalatrial fibrillation: The effect of digoxin. Br Heart J1990; 63:225-227.

2. Galun E, Flugelman MY, Glickson M, et al. Failureof long-term digitalization to prevent rapid ventric-ular response in patients wilh paroxysmal atrial fih-rillation. Chest 1991; 99(4):1038-1040.

3. Falk RH. Leavitt JI. Digoxin for atrial fihrillation: Adrug whose time has gone? Ann Intern Med 1991;114:573-575.

4. Murgatroyd FD. O'Nunain S, Gihson SM, ct al.CRAFT Investigators: The results of CRAFT-1: Amulti-center, double-blind, placeho-controlledcrossover study of digoxin in symptomatic paroxys-mal atrial fihrillation, (ahstract) J Am Coll Cardiol1993; 21:478A.

Murgatroyd FD, Xie B, Copie X. et al. Identificationof atrial fihrillation episodes in ambulatory electro-cardiographic recordings; Validation of a methodfor ohtaining lahelled R-R interval files. PACE 1995;18:1315-1320.Geraets DR, Kienzle MG. Atrial fihrillation andatrial flutter. Clinical Pharmacy 1993; 12:721-735.Falk RH, Knowhon AA, Bernard. SA, ct al. Digoxinfor converting recent-onset atrial fihrillalion to si-nus rhythm. A randomizfid, double blind trial. AnnhUern Med 1987; 106:503-505.Gonzalez ER, Ornato JP, l.awson GL. Clinical deci-sion analysi.s modelling; Short-tRrm control of ven-tricular response rate in atrial fihrillation or atrialflutter-digoxin versus diltiazem. Fharmacotherapy1994; 14(4):44B^51.

1740 September 1998 PACE, Vol. 21