Lack of evidence of increased mortality among patients with atrial fibrillation taking digoxin: findings from post hoc propensity-matched analysis of the AFFIRM trial

We thank Dr. Elayi and his colleague for their interest and fully agree that there were methodological differences between our two analyses.1, 2 We have clearly noted those differences in our paper and they were further highlighted in an accompanying statistical viewpoint.3

We believe that the method of using digoxin as a "time- dependent" treatment is problematic as digoxin is often prescribed for heart failure, a severe condition with high morbidity and mortality. Three -quarters of all AFFIRM patients with a history of heart failure were receiving digoxin at baseline.

According to Fisher and Lin, when patients receive a completely ineffective treatment "only when a certain intermediate event occurs" that may increase the risk of death then "this is the effect of the intermediate event, not the real effect of the treatment" and "such an analysis would be completely misleading."4

They argue that the use of "time-dependent" treatment method in this scenario would violate a fundamental assumption that the "change in exposure or treatment occurs in a random fashion", and because it may be "difficult to adjust for the factors that trigger the change through modeling", it is important to be "extremely cautious when interpreting results involving time-dependent exposure or treatment."4

The success of propensity score method, on the other hand, is "often judged by whether balance on covariates is achieved between experimental and control group after its use."5 Patients receiving and not receiving digoxin in our outcome-blinded propensity-matched cohort were balanced on 59 baseline characteristics.2

Although exclusion of patients from matched cohorts may limit external validity, it strengthens internal validity, which by definition, precedes external validity.6 As noted in the accompanying statistical viewpoint, "similar results were observed in a sensitivity analysis using propensity adjustment rather than propensity matching, which allowed for inclusion of the larger cohort (n=2706)."3

Both analyses may be limited by bias due to unmeasured confounders, and both are limited by inherent weaknesses of the AFFIRM dataset. Randomized controlled trials of digoxin in chronic heart failure without atrial fibrillation have demonstrated no increase in mortality.7 Currently there is no evidence to suggest that the effect of digoxin on mortality might vary based on the presence or absence of atrial fibrillation.

Of note, in AFFIRM, digoxin was one of the rate-control drugs and the relative risk of all-cause death was non-significantly 13% (p=0.08) lower in the rate control group overall, and significantly 20% (p=0.009) lower among those 65 years or older.8

Digoxin is recommended for rate control by major national atrial fibrillation guidelines.9, 10 In conclusion, we would like to reiterate that we found no evidence of an increased risk of mortality or hospitalization among patients receiving digoxin in AFFIRM and that based on existing data there is no need to reassess the role of digoxin in the management of atrial fibrillation.

Mihai Gheorghiade, Gregg C. Fonarow, Wilbert S. Aronow, Stefan D. Anker and Ali Ahmed

References:

1. Whitbeck MG, Charnigo RJ, Khairy P, Khaled Ziada, Alison L. Bailey, Milagros M. Zegarra, Jignesh Shah, Gustavo Morales, Tracy Macaulay, Vincent L. Sorrell, Charles L. Campbell, John Gurley, Paul Anaya, Hafez Nasr, Rong Bai, Luigi Di Biase, David C. Booth, Guillaume Jondeau, Andrea Natale, Denis Roy, Susan Smyth, David J. Moliterno, Elayi CS. Increased mortality among patients taking digoxin analysis from the AFFIRM study. Eur Heart J 2012; doi:10.1093/eurheartj/ehs348.

2. Gheorghiade M, Fonarow GC, van Veldhuisen DJ, Cleland JGF, Butler J, Epstein AE, Patel K, Aban IB, Aronow WS, Anker SD and Ahmed A. Lack of evidence of increased mortality among patients with atrial fibrillation taking digoxin: findings from post hoc propensity-matched analysis of the AFFIRM trial Eur Heart J 2013; doi: 10.1093/eurheartj/eht120 First published online: April 16.

3. Murphy SA. When 'digoxin use' is not the same as 'digoxin use': lessons from the AFFIRM trial. Eur Heart J 2013; doi: 10.1093/eurheartj/eht087 First published online: April 16.

4. Fisher LD, Lin DY. Time-dependent covariates in the Cox proportional-hazards regression model. Annu Rev Public Health 1999; 20:145 -157.

5. Heinze G, Juni P. An overview of the objectives of and the approaches to propensity score analyses. Eur Heart J 2011; 32:1704-1708.

6. Curran PJ, Wirth RJ. Interindividual Differences in Intraindividual Variation: Balancing Internal and External Validity. Measurement 2004; 2:219-247.

7. The Digitalis Investigation Group Investigators. The effect of digoxin on mortality and morbidity in patients with heart failure. N Engl J Med 1997; 336:525-533.

8. Wyse DG, Waldo AL, DiMarco JP, Domanski MJ, Rosenberg Y, Schron EB, Kellen JC, Greene HL, Mickel MC, Dalquist JE, Corley SD, Atrial Fibrillation Follow-up Investigation of Rhythm Management I. A comparison of rate control and rhythm control in patients with atrial fibrillation. N Engl J Med 2002; 347:1825-1833.

9. Camm AJ, Kirchhof P, Lip GY, Schotten U, Savelieva I, Ernst S, Van Gelder IC, Al-Attar N, Hindricks G, Prendergast B, Heidbuchel H, Alfieri O, Angelini A, Atar D, Colonna P, De Caterina R, De Sutter J, Goette A, Gorenek B, Heldal M, Hohloser SH, Kolh P, Le Heuzey JY, Ponikowski P, Rutten FH, European Heart Rhythm A, European Association for Cardio- Thoracic S. Guidelines for the management of atrial fibrillation: the Task Force for the Management of Atrial Fibrillation of the European Society of Cardiology (ESC). Eur Heart J 2010; 31:2369-2429.

10. Fuster V, Ryden LE, Cannom DS, Crijns HJ, Curtis AB, Ellenbogen KA, Halperin JL, Kay GN, Le Huezey JY, Lowe JE, Olsson SB, Prystowsky EN, Tamargo JL, Wann LS, Smith SC, Jr., Priori SG, Estes NA, 3rd, Ezekowitz MD, Jackman WM, January CT, Page RL, Slotwiner DJ, Stevenson WG, Tracy CM, Jacobs AK, Anderson JL, Albert N, Buller CE, Creager MA, Ettinger SM, Guyton RA, Hochman JS, Kushner FG, Ohman EM, Tarkington LG, Yancy CW, American College of Cardiology Foundation/American Heart Association Task F. 2011 ACCF/AHA/HRS focused updates incorporated into the ACC/AHA/ESC 2006 guidelines for the management of patients with atrial fibrillation: a report of the American College of Cardiology Foundation/American Heart Association Task Force on practice guidelines. Circulation 2011; 123:e269- 367.

Conflict of Interest:

None declared

We read with great interest Gheorghiade et al.'s AFFIRM analysis on digoxin use in patients with AF [1]. Whereas we recently reported a statistically significant estimated hazard ratio (HR) of 1.41 for increased mortality with digoxin (95% confidence interval [CI; 1.19-1.67]) using the same dataset [2], Gheorghiade et al. found no such association (HR 1.06, 95% CI [0.83-1.37]). There is little doubt that these disparate results reflect methodological differences.

In our propensity-adjusted model, digoxin use was considered a time- dependent variable, meaning that changes in therapy were accounted for by updating analyses, such that time at risk for death was attributed to the actual, as opposed to assumed, exposure category. Such an analysis carries the potential to generate more accurate results by reducing misclassification errors and optimizing available data. However, as noted by Gheorghiade et al., caution must be exercised when modeling time- dependent treatments if the change in therapy is related to health status. For example, bias may be introduced if changes in heart failure status prompted decisions regarding digoxin therapy. To minimize such a bias, we modeled New York Heart Association (NYHA) functional class as a time- dependent variable. If changes in heart failure status were reflected by changes in NYHA functional class, this potential bias would be eliminated. As further protection against this bias, we likewise modeled Canadian Cardiovascular Society angina status, heart rate, permanency of AF, beta- blocker use, angiotensin-converting enzyme inhibitor use, amiodarone use, and number of cardioversions as time-dependent variables and adjusted for baseline heart failure status. Moreover, a subgroup analysis of patients with no changes to therapy (i.e., "on" or "off" digoxin throughout the study) yielded a similar conclusion [HR 1.58, 95% CI (1.23-2.03)]. It is, therefore, implausible that the association between digoxin and mortality may be accounted for by so-called indication bias.

In contrast, Gheorghiade et al.'s analysis essentially consisted of propensity matching a subset of qualifying patients on the basis of digoxin use assessed at a single time-point. Such an approach may be preferred if patients selected for inclusion are not systematically different from those excluded, matching does not result in a substantial loss of study power, and the exposure defined at baseline remains constant over time. Unfortunately, none of these conditions hold true in the present instance. Gheorghiade et al.'s main analyses are based on 43.3% of the original data set (878 pairs of patients). As acknowledged by the authors, this highly selected subset of patients was systematically different from those excluded, producing the potential for a major selection bias at entry. Indeed, the primary outcome, i.e., mortality, was significantly higher among the 1352 patients discarded prior to propensity matching than among the 2706 patients retained for analysis (P<0.001). The sensitivity analysis, though helpful, nevertheless excluded 1152 (28.4%) patients. Perhaps most importantly, the critical assumption that the effect of exposure defined at baseline remained constant over time was conspicuously violated. For example, among the 2153 patients who received digoxin within 6 months preceding randomization, 55.8% discontinued therapy during the study. Among the 1905 patients not receiving digoxin within 6 months preceding randomization, 34.9% received digoxin at a later date. The sheer magnitude of this bidirectional "crossover" renders an analysis based on a single assessment of digoxin status underpowered at best and meaningless at worst [3].

Variations in methodological assessment of the relationship between digoxin and mortality reflect the fact that different approaches may be used to minimize biases inherent to observational research. However, the worrisome signal linking digoxin to increased mortality has been identified by various studies employing different designs [4]. For example, a cohort study of over 23,000 patients with AF and no heart failure recently reported that digoxin incurred a twofold-increased risk of mortality (HR 2.06, 95% CI [1.73-2.45]) [5]. In our opinion, Gheorghiade et al.'s analysis, which is subject to potentially major selection and misclassification biases towards the null hypothesis, offers little reassurance. Pending more definitive data from prospective randomized controlled trials, it would seem cautious to discourage widespread adoption of a rate control strategy for AF that favours digoxin as a single first line agent, especially when other safe and inexpensive alternatives such as beta blockers or calcium blockers are readily available [6].

[1] Gheorghiade M, Fonarow G, Van Veldhuisen D, Cleland J, Butler J, Epstein A, Patel K, Aban I, Aronow W, Anker S, Ahmed A: Lack of evidence of increased mortality among patients with atrial fibrillation taking digoxin: findings from post hoc propensity-matched analysis of the AFFIRM European Heart Journal (in press) doi:10.1093/eurheartj/eht120.

[2] Whitbeck MG, Charnigo RJ, Khairy P, Ziada K, Bailey AL, Zegarra MM, Shah J, Morales G, Macaulay T, Sorrell VL, Campbell CL, Gurley J, Anaya P, Nasr H, Bai R, Di Biase L, Booth DC, Jondeau G, Natale A, Roy D, Smyth S, Moliterno DJ, Elayi CS. Increased mortality among patients taking digoxin-analysis from the AFFIRM study. Eur Heart J. 2012 Nov 27. [Epub ahead of print].

[3] D'Amico R, Petracci E, Balduzzi S, Moja L, Miglio R. Bias associated with selective crossover in randomized controlled trials Oct. 2012 Abstract (http://2012.colloquium.cochrane.org/abstracts/bias- associated-selective-crossover-randomised-controlled-trials).

[4] Hallberg P, Lindback J, Lindahl B, Stenestrand U, Melhus H; RIKS- HIA group. Digoxin and mortality in atrial fibrillation: a prospective cohort study. Eur J Clin Pharmacol 2007;63:1201-1202.

[5] Freeman J; Reynolds K; Fang M; Udaltsova N; Steimle A; Singer D; Go A: Digoxin and risk of death in adults with atrial fibrillation: the ATRIA-CVRN STUDY J Am Coll Cardiol. 2013;61(10_S): doi:10.1016/S0735- 1097(13)60262-4.

[6] Whitbeck M, Khairy P, Elayi CS: Response to Dr Horowitz's letter EHJ 26 December 2012.

Conflict of Interest:

None declared