OUP user menu

Mixed treatment comparison of dronedarone, amiodarone, sotalol, flecainide, and propafenone, for the management of atrial fibrillation

Nick Freemantle, Carmelo Lafuente-Lafuente, Stephen Mitchell, Laurent Eckert, Matthew Reynolds
DOI: http://dx.doi.org/10.1093/europace/euq450 329-345 First published online: 12 January 2011


Aims Mixed treatment comparisons (MTC) were performed to assess the relative efficacy and tolerability of the main anti-arrhythmic drugs used for the treatment of atrial fibrillation (AF)/flutter.

Methods and results Electronic databases were systematically searched to identify randomized controlled trials (RCTs) examining amiodarone, dronedarone, flecainide, propafenone, sotalol, or placebo for the treatment of AF. Thirty-nine RCTs met inclusion criteria and were combined using MTC models to provide direct and indirect comparisons in a single analysis. Results are presented vs. placebo. Amiodarone had the largest effect in reducing AF recurrence (OR 0.22, 95% CI 0.16–0.29). Amiodarone was associated with the highest rate of patients experiencing at least one serious adverse event (OR 2.41, 95% CI 0.96–6.06) and treatment withdrawals due to adverse events (OR 2.91, 95% CI 1.66–5.11). Dronedarone was associated with the lowest rate of proarrhythmic events including bradycardia (OR 1.45, 95% CI 1.02–2.08). Dronedarone significantly reduced the risk of stroke (OR 0.69, 95% CI 0.57–0.84). Trends towards increased mortality for sotalol (OR 3.44, 95% CI 1.02–11.59) and amiodarone (OR 2.17, 95% CI 0.63–7.51) were found, which were stronger when small studies randomizing <100 subjects per group were excluded.

Conclusions Amiodarone has been demonstrated to be the most effective drug in maintaining sinus rhythm. Differences in outcomes between the anti-antiarrhythmic drugs were reported, with sotalol and possibly amiodarone increasing mortality and dronedarone possibly decreasing the incidence of serious adverse events and proarrhythmia.


Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia.13 It is a progressive disease and is strongly associated with adverse clinical outcomes, including heart failure and stroke, resulting in considerable morbidity and mortality.46

Despite efforts at both rhythm and rate control, patients with AF and atrial flutter (AFL) are at a markedly increased risk of cardiovascular hospitalizations and cardiovascular death, particularly in older age groups. Atrial fibrillation is the most frequent arrhythmic cause of hospital admission in the USA, representing more than one-third of all patient discharges with arrhythmia as a principal diagnosis.7

Pharmacological treatments for AF were reviewed in 2007 by Lafuente-Lafuente et al.8 Endpoints evaluated included all-cause mortality, AF recurrences, and withdrawals due to adverse events (AEs) pro-arrhythmia and embolic complications. The authors concluded that amiodarone was the most effective drug in maintenance of sinus rhythm (SR), but the usefulness of this agent may be limited by toxicity. In 2009, dronedarone, a new anti-arrhythmic drug (AAD) with multiple ion channel blockade properties demonstrated benefit in the composite outcome of cardiovascular hospitalization or death from all causes in the treatment of AF.9

To date, there are limited data directly describing the safety and effectiveness of dronedarone compared with alternative AADs in AF patients. In the absence of direct comparisons, there has been increasing interest in undertaking mixed treatment comparisons (MTCs) of networks of trials in order to provide best estimates of the relative effectiveness and safety of alternative health-care interventions.10 Although these analyses provide a lesser strength of evidence than well designed and adequately powered directly randomized trials, they do provide a summary of the best available evidence, borrowing weight from indirect comparisons.

In this paper, we present the results of MTCs for amiodarone, dronedarone, flecainide, propafenone, and sotalol, for the effect of treatment on outcomes of all-cause mortality, stroke, prevention of AF recurrence, withdrawals (all cause and specifically due to AE), serious adverse events (SAEs), and incidence of proarrhythmic events. We also provide conventional meta-analyses of direct comparisons.


Inclusion/exclusion criteria and search strategy

To identify relevant studies, electronic databases and conference proceedings were searched: Medline, EMBASE, and the Cochrane Central Register of Controlled Trials were accessed on 8 April 2009. There were no restrictions by date of publication. The search combined both index and free-text terms for ‘AF/flutter’ with the interventions ‘AADs’, pharmacological therapies, ‘ablation’ and publication type ‘prospective clinical study’ or ‘randomized clinical trial’ (RCT), or studies reporting quality-of-life outcomes.

Only RCTs, which examined the effect of amiodarone, dronedarone, flecainide, propafenone, or sotalol in patients with AF or AFL, were included. The class IA AADs (e.g. quinidine, procainamide, disopyramide) were not included as part of the search strategy because a previous synthesis of evidence concluded that they are potentially harmful in AF patients,8 and they are no longer recommended in consensus guidelines.11 Randomized controlled trials which compared the target treatments with an inactive control (placebo or no treatment) or with each other were included. Studies enrolling patients with AF/AFL ≤3 months after cardiac surgery, with follow-up ≤3 months, or with duration of treatment <30 days were excluded. Identified studies were assessed independently by two reviewers in order to ascertain whether they met the pre-defined inclusion/exclusion criteria, and discrepancies were resolved by a third party. Data were extracted from included publications by a reviewer into an Excel® spreadsheet. A second reviewer checked the resulting extraction and any discrepancies were resolved through discussion.

The methodological quality of the included RCTs was abstracted and assessed according to methods recommended in section six of the Cochrane Reviewer's handbook version In brief, the likelihood of bias was assessed according to three components: adequacy of randomization and allocation concealment procedures, adequacy of blinding procedures and completeness of follow-up. Descriptors of the trial populations and interventions used were also abstracted.


Mortality, stroke, AF recurrence, incidence of SAEs, treatment withdrawals (all cause and due specifically to AEs) and proarrhythmic events were analysed. For these analyses, AF recurrence was defined as the reported number of patients failing to maintain SR at any point within the study timeframe (SR maintenance data), or the reported number of patients with a recurrence of AF within the timeframe of the study (AF recurrence data).

Serious adverse events were defined as those events reported to be serious in the publication by the investigator. It was not possible to discriminate between SAEs that were reported using a standard regulatory definition and other SAEs.

Proarrhythmic events were reported as per the Cochrane analysis8 and included sudden death, new symptomatic arrhythmia, bradycardia, and drug discontinuation due to new QRS or QT interval prolongation.13

Statistical methods

Placebo control arms were pooled with no treatment/untreated control arms. Trials with multiple arms of different doses of the same drug were pooled into a single-treatment arm prior to meta-analysis across trials.

Direct meta-analysis of included trials was estimated using random effect models and the Peto odds ratio (OR).14 Data were analysed using an intention-to-treat (ITT) approach, and were based on all data on patients allocated to an intervention, including patients who withdrew from the trial and those who did not complete treatment. The analysis of SAE and treatment discontinuations included data reported at the end of the study whatever the follow-up duration. These analyses were performed using STATA, version 10.1.

Random effects MTC models were developed which combined direct and indirect information from clinical trials appropriately conditioning parameter variances through adding a multiplicative over dispersion parameter on the trial stratum. The data analysis for this paper was generated using SAS software. Copyright, SAS Institute, Inc., SAS and all other SAS Institute, Inc., product or service names are registered trademarks or trademarks of SAS Institute, Inc., Cary, NC, USA. The SAS code used to specify these models is described in the Appendix.

Bayesian methods have increasingly been used for indirect comparisons. However, these models are known to rely heavily on the prior information when event rates are sparse.15 As this was anticipated due to the particular nature of trials that were to be included, non-linear mixed models based upon pseudo-likelihood were preferred. The latter models derive their estimations solely from the data. In general, the non-linear mixed models were qualitatively very similar to those unconstrained Bayesian baseline models specified using numerical simulation techniques and described by Lu and Ades.10

In line with standard methods for conventional meta-analyses, we excluded studies which reported no events in all groups from the meta-analysis for each respective endpoint.16,17 The between trial heterogeneity derived from the covariance statistic and standard error on that stratum was reported. When heterogeneity was present, additional analyses excluding small studies (<100 patients in either arm with at least one event in at least one arm) were conducted to explore potential causes. Results are reported against placebo in each case and described using Forest plots.


In total, 10 743 references were identified through electronic database searching, of which 9322 were excluded on the basis of title and abstract (Figure 1). On re-application of the review inclusion criteria to the 1411 full-text papers, 1299 were excluded for design and/or comparator considerations (Figure 1).

Figure 1

Diagram flow. *Number of studies that reported the event and at least 1 event was observed during follow-up.

Of the 113 remaining publications, 39 met the inclusion criteria, reported one of the outcomes of interest, and compared the target treatments with an inactive control (placebo or no treatment) or with each other. Data from the DIONYSOS trial were taken from the FDA dossier;18 final results were recently published.19 The analyses were conducted using these 40 publications. Overall there was some variation in the quality of trials included in the analysis data set (Appendix Table A1). The baseline characteristics of the data set analysed are presented in Table 1.

View this table:
Table 1

Baseline characteristics summary of the included studies

Patient characteristicsRange of valuesMean (SD)
Number of patients randomized16–4628245 (580)
Age (years)49–7861.6 (5.2)
Gender (% male)35–9959.2 (12.6)
AF Type (paroxysmal)a33
AF type (persistent)a23
AF type (permanent)a7
Structural heart disease (%)0–10060.1 (23.4)
Left ventricular ejection fraction (LVEF) (%)30–6855.1 (9.1)
Left atrium diameter (LAD) (mm)34–5042.9 (3.8)
  • aNumber of studies reported across the included studies.

The number of patients randomized per study varied from 16 to 4628. The mean age of subjects across studies was 62 years and 59% of subjects were male. The majority of studies included patients with paroxysmal AF. Permanent AF patients were included in addition to other types of AF in seven of the studies. A majority of patients had structural heart disease, mean left ventricular ejection fraction (LVEF) was 55% and left atrial diameter (LAD) at echocardiography was 42.9 mm.

For the purposes of contrast with the MTC, conventional direct estimates of treatment effect are described in Tables 2 and 3. The MTC analyses used the full network of trials to estimate the comparison with placebo and not only the trials that compared a particular treatment with placebo.

View this table:
Table 2

Direct meta-analysis: efficacy summary results

ComparisonMortalityStrokeAF recurrence
No. of studiesn/NOR (95% CI)No. of studiesn/NOR (95% CI)No. of studiesn/NOR (95% CI)a
Dronedarone vs.
 Placebo4136/3418 vs. 142/28910.86 (0.67–1.10)250/3129 vs. 70/27360.69 (0.47–0.99)2700/1032 vs. 369/4750.59 (0.45–0.76)
 Amiodarone12/249 vs. 5/2550.40 (0.08–2.13)12/249 vs. 2/2551.02 (0.14–7.33)1158/249 vs. 107/2552.38 (1.67–3.45)
Amiodarone vs.
 Placebo613/472 vs. 5/2991.08 (0.12–9.42)16/267 vs. 3/1371.03 (0.25–4.17)6170/472 vs. 232/2990.15 (0.10–0.22)
 Sotalol428/490 vs. 39/4830.66 (0.40–1.11)16/267 vs. 5/2611.18 (0.35–3.90)5185/490 vs. 269/5180.47 (0.36–0.62)
 Flecainide231/100 vs. 34/1040.74 (0.40–1.38)
 Propafenone10/72 vs. 0/741.55 (0.26–9.17)129/72 vs. 35/740.75 (0.39–1.45)b
Sotalol vs.
 Placebo1227/1686 vs. 5/10922.52 (0.96–6.64)15/261 vs. 3/1370.87 (0.21–3.71)10861/1424 vs. 677/10470.41 (0.28–0.59)
 Flecainide10/20 vs. 0/201.00 (0.06–16.50)18/20 vs. 6/201.56 (0.42–5.76)
 Propafenone32/241 vs. 0/2385.20 (0.24–111.24)4138/275 vs. 155/2770.76 (0.52–1.09)
Flecainide vs.
 Placebo42/99 vs. 0/1055.24 (0.24–112.45)330/71 vs. 56/780.25 (0.09–0.70)
 Propafenone20/145 vs. 1/1522.29 (0.64–8.16)12/48 vs. 1/492.09 (0.18–23.78)130/97 vs. 30/1031.09 (0.59–2.00)
Propafenone vs.
 Placebo72/945 vs. 3/4920.42 (0.08–2.05)6454/944 vs. 349/4810.29 (0.16–0.55)
  • aOR for 12 month timepoint.

  • b24 months data.

View this table:
Table 3

Direct meta-analysis: safety summary results

Treatment withdrawalsTreatment withdrawals due to AESerious adverse eventsPro-arrhythmia events
ComparisonStudiesn/NOR (95% CI)Studiesn/NOR (95% CI)Studiesn/NOR (95% CI)Studiesn/NOR (95% CI)
Dronedarone vs.
 Placebo4883/3418 vs. 787/28911.23 (0.85–1.77)4409/3418 vs. 226/28911.63 (1.32–2.03)3492/2580 vs. 501/2468a1.22 (0.59–2.22)4244/3418 vs. 153/28911.46 (1.18–1.80)
 Amiodarone196/249 vs. 69/2551.69 (1.16–2.44)132/249 vs. 45/2550.69 (0.43–1.12)132/249 vs. 45/2550.69 (0.43–1.12)12/249 vs. 11/2550.29 (0.10–0.90)
Amiodarone vs.
 Placebo670/472 vs. 29/2993.92 (0.64–24.02)528/217 vs. 1/1958.41 (2.17–32.63)111/65 vs. 0/608.10 (2.36–27.81)513/445 vs. 2/2682.27 (0.74–6.99)
 Sotalol580/490 vs. 70/5181.12 (0.58–2.20)438/281 vs. 31/2671.13 (0.35–3.68)213/100 vs. 6/961.80 (0.31–10.42)413/521 vs. 22/4920.56 (0.28–1.12)
 Flecainide28/100 vs. 6/1041.81 (0.56–5.83)
 Propafenone117/72 vs. 4/744.34 (1.73–10.91)117/72 vs. 4/744.34 (1.73–10.91)117/72 vs. 2/746.26 (2.39–16.37)12/72 vs. 2/741.03 (0.13–7.45)
Sotalol vs.
 Placebo10712/1424 vs. 416/10470.91 (0.62–1.32)10238/1424 vs. 96/10472.06 (1.18–3.62)6222/1482 vs. 96/9211.77 (0.95–1.30)1168/1650 vs. 12/10612.21 (1.21–4.01)
 Flecainide10/20 vs. 0/201 (0.06–16.5)10/20 vs. 0/201 (0.06–16.5)15/20 vs. 3/201.83 (0.39–8.49)
 Propafenone426/275 vs. 20/2771.32 (0.72–2.43)426/275 vs. 20/2771.32 (0.72–2.43)29/132 vs. 7/1271.34 (0.48–3.78)428/279 vs. 20/2791.47 (0.76–2.84)
Flecainide vs.
 Placebo37/71 vs. 0/782.92 (0.22–38.62)37/71 vs. 0/788.45 (1.02–70.40)29/79 vs. 0/8010.36 (1.26–58.24)415/114 vs. 1/1216.60 (1.52–28.58)
 Propafenone248/145 vs. 56/1520.80 (0.40–1.61)212/145 vs. 18/1520.54 (0.09–3.23)12/48 vs. 9/490.25 (0.07–0.86)23/145 vs. 6/1520.60 (0.05–6.58)
Propafenone vs.
 Placebo7155/955 vs. 54/4921.55 (1.08–2.22)7123/955 vs. 27/4922.00 (1.22–3.30)474/493 vs. 21/3202.12 (0.80–5.60)837/1067 vs. 12/6042.34 (0.76–7.27)
  • aWithout EURIDIS/ADONIS.

Direct meta-analyses

In the direct meta-analysis, there were few mortality and stroke events leading to uncertainty in the estimation for these analyses. None of the studies that investigated flecainide or propafenone reported stroke outcomes. Dronedarone had the highest number of patients investigated primarily due to the inclusion of the ATHENA trial. A large number of subjects in the trials experienced the outcome of AF recurrence, providing a good basis for the estimation of the effect sizes.

Sotalol showed a trend towards an increased risk of mortality compared with placebo [OR 2.52, 95% confidence interval (CI) 0.96–6.64]. All other comparisons were not statistically significant. Dronedarone was the only drug to show a statistically significant reduction in the risk of stroke. The estimated OR was 0.69, 95% CI 0.47–0.99.

All drugs were shown to be efficacious at reducing AF recurrence. Amiodarone was associated with the highest efficacy with an OR of 0.15, 95% CI 0.10–0.22. Furthermore, amiodarone was statistically superior to both sotalol (OR 0.47, 95% CI 0.36–0.69) and dronedarone (OR 0.42, 95% CI 0.29–0.60) in direct comparisons.

Safety was investigated through four outcomes: treatment withdrawals for any reason; and treatment withdrawals due to AEs, SAEs, and proarrhythmia. Studies investigating dronedarone, sotalol, or propafenone enrolled >1000 patients compared with those for amiodarone and flecainide which recruited only 472 and 114 patients, respectively.

On the outcome of treatment withdrawals for any reason, propafenone was the only AAD to show a significant increase vs. placebo. Dronedarone treatment significantly increased treatment withdrawals for any reason compared with amiodarone (OR 1.69, 95% CI 1.16–2.44). Similarly, amiodarone significantly increased treatment withdrawals for any reason compared with propafenone (OR 4.34, 95% CI 1.73–10.91).

Treatment withdrawals specifically due to AEs were significantly increased for all AADs compared with placebo. Dronedarone presented the lowest increase in the risk of events with an OR of 1.63, 95% CI 1.32–2.03. The highest OR was reported for amiodarone compared with placebo where the odds of withdrawal were estimated at 8.41, 95% CI 2.17–32.63. Amiodarone treatment resulted in significantly increased treatment withdrawals due to AEs when compared with propafenone (OR 4.34, 95% CI 1.73–10.91).

Based on a limited number of patients both amiodarone and flecainide showed an increase in the risk of SAEs compared with placebo; the ORs were 8.10, 95% CI 2.36–27.81 and 10.36, 95% CI 1.26–58.24, respectively. The OR for SAEs for sotalol compared with placebo was estimated at 1.77, 95% CI 0.95–1.30. Amiodarone significantly increased the risk of SAEs compared with propafenone with an OR of 6.26, 95% CI 2.39–16.37. All drugs were associated with an increased risk of proarrhythmia compared with placebo. The increase did not reach statistical significance for amiodarone OR: 2.27, 95% CI 0.74–6.99 or propafenone (OR: 2.34, 95% CI 0.76–7.27). However, the risk of proarrhythmia was statistically higher with amiodarone when compared with dronedarone (OR: 3.41, 95% CI 1.11–10.44).

Mixed treatment comparison analysis: all-cause mortality

In total, 18 trials9,18,2035 reported mortality and included at least one event in at least one relevant randomized group. In these trials, 10 032 patients were randomized and 369 patients died. Trials included 3614 patients randomized to dronedarone, 680 patients randomized to amiodarone, 1342 patients randomized to sotalol, 468 patients randomized to propafenone, 90 patients randomized to flecainide, and 3838 patients randomized to placebo. The effect of each AAD compared with placebo on all-cause mortality is described in Figure 2.

Figure 2

Mixed treatment comparison analysis: effect of anti-arrhythmic drugs on all-cause mortality. Odds ratios and 95% confidence intervals. Note—odds ratio smaller than 1 indicates a benefit (lower mortality) for the active agent.

Sotalol was associated with an increase in risk of mortality compared with placebo/control (P = 0.046). The analysis contained several small trials which introduced substantial uncertainty—the between trials covariance parameter estimate was 1.71 (SE 0.63; P = 0.006). This leads to wide CIs and imprecision in the between agent measures. A further sensitivity analysis restricted the sample to those trials comparing the target drug either with an untreated control condition or an alternative drug, with at least 100 patients per randomized group and at least one event in either group. In total 7 trials,9,18,20,26,29,31,32 were included, in which 8252 patients were randomized and in 349 patients died. Trials included 3378 patients randomized to dronedarone, 653 patients randomized to amiodarone, 873 patients randomized to sotalol, and 3348 patients randomized to placebo. The between trials covariance parameter estimate was reduced to 0.5860 (SE 0.3706; P = 0.12).

The effect of each AAD compared with placebo on all-cause mortality in this subset of larger trials is described in Figure 3. Both amiodarone and sotalol were associated with an increased risk of death in this restricted analysis.

Figure 3

Mixed treatment comparison analysis: effect of anti-arrhythmic drugs on all-cause mortality in studies involving >100 patients in either arm. Odds ratios and 95% confidence intervals. Note—odds ratio smaller than 1 indicates a benefit (lower mortality) for the active agent.

Mixed treatment comparison analysis: stroke

In total, 4 trials were included9,18,31,32 in which 7034 patients were randomized and 138 patients experienced a stroke. Trials included 261 patients randomized to sotalol, 3378 patients randomized to dronedarone, 522 patients randomized to amiodarone, and 2873 patients randomized to placebo. The effect of each AAD compared with placebo on stroke is described in Figure 4.

Figure 4

Mixed treatment comparison analysis: effect of anti-arrhythmic drugs on stroke. Odds ratios and 95% confidence intervals. Note—odds ratio lower than 1 describes a lower rate of stroke for the active treatment.

Randomization to dronedarone was associated with statistically significant reductions in stroke compared with control. Neither amiodarone nor sotalol achieved statistically significant reductions in stroke compared with placebo.

There was no evidence of heterogeneity between studies (covariance parameter 0.06, SE0.06, P = 0.3173).

Mixed treatment comparison analysis: atrial fibrillation recurrence

In total, 30 trials were included18,20,21,24,25,27,2952 in which 6629 patients were randomized and 3775 patients experienced AF recurrence. Trials included 1228 patients randomized to propafenone, 1404 patients randomized to sotalol, 1131 patients randomized to dronedarone, 978 patients randomized to amiodarone, 305 patients randomized to flecainide, and 1583 patients randomized to placebo. The effects of each AAD compared with placebo on AF recurrence are described in Figure 5.

Figure 5

Mixed treatment comparison analysis: effect of anti-arrhythmic drugs on atrial fibrillation recurrence. Odds ratios and 95% confidence intervals. Note—odds ratio lower than 1 describes a lower rate of atrial fibrillation recurrence for the active treatment.

All treatments showed statistically significant reductions in AF recurrence compared with placebo. There was substantial between study variability (covariance parameter 1.59, SE0.42, P = 0.0002). Limiting the analysis to trials including at least 100 patients per group reduced heterogeneity (P = 0.11), and led to the exclusion of flecainide from the analysis, but otherwise had no qualitative effect upon the results.

Mixed treatment comparison analysis: treatment discontinuation

In total, 34 trials were included9,18,21,2328,3036,3949,5156 in which 12 239 patients were randomized and 3150 patients discontinued. Trials included 3667 patients randomized to dronedarone, 1416 patients randomized to propafenone, 1302 patients randomized to sotalol, 1172 patients randomized to amiodarone, 429 patients randomized to flecainide, and 4253 patients randomized to placebo. The effect of each AAD compared with placebo on treatment discontinuation is described in Figure 6.

Figure 6

Mixed treatment comparison analysis: effect of anti-arrhythmic drugs on treatment discontinuation. Odds ratios and 95% confidence intervals. Note—odds ratio smaller than 1 indicates a benefit (lower discontinuation) for the active agent.

Numerically sotalol was associated with the lowest rate of treatment discontinuation. However, there were no statistically significant differences in tolerability between the agents compared with placebo. There was substantial heterogeneity between studies (between study covariance parameter 2.96, SE0.70, P < 0.0001). Limiting the analysis to trials including at least 100 patients in each group reduced between study variability (covariance parameter 1.03, SE0.65, P = 0.11), but did not affect the results of the analysis.

Mixed treatment comparison analysis: withdrawal due to adverse events

In total, 29 trials were included9,18,20,21,2330,3236,3946,49,51,52,54,57 in which 11 763 patients were randomized and 1342 patients experienced at least one SAE. Trials included 1624 patients randomized to sotalol, 3667 patients randomized to dronedarone, 716 patients randomized to amiodarone, 224 patients randomized to flecainide, 1261 patients randomized to propafenone, and 4271 patients randomized to placebo. The effect of each AAD compared with placebo on the incidence of withdrawal due to AEs is described in Figure 7.

Figure 7

Mixed treatment comparison analysis: effect of anti-arrhythmic drugs on withdrawal due to adverse events. Odds ratios and 95% confidence intervals.

Treatment with amiodarone, dronedarone, or sotalol was associated with significantly increased odds of withdrawals due to AEs compared with placebo. There was considerable evidence of between study heterogeneity (covariance parameter 2.36, SE0.63, P = 0.0002). Limiting the analysis to those patients with at least 100 patients in each group reduces between study heterogeneity (P = 0.16) but had no qualitative effect upon the results.

Mixed treatment comparison analysis: serious adverse events

In total, 20 trials were included9,18,21,23,2529,3234,4246,51,52 in which 9734 patients were randomized and 1716 patients experienced at least one SAE. Trials included 1113 patients randomized to sotalol, 3657 patients randomized to dronedarone, 427 patients randomized to amiodarone, 231 patients randomized to flecainide, 550 patients randomized to propafenone, and 3756 patients randomized to placebo. The effect of each agent compared with placebo on the incidence of SAEs is described in Figure 8.

Figure 8

Mixed treatment comparison analysis: effect of anti-arrhythmic drugs on incidence of serious adverse events. Odds ratios and 95% confidence intervals. Note—odds ratio lower than 1 describes a lower rate of serious adverse events for the active treatment.

No agent was statistically different from placebo on the incidence of SAEs. There was some evidence of between trial heterogeneity (covariance parameter 3.83, SE1.35, P = 0.005). Limiting the trials to those with at least 100 patients in each group resulted in homogeneity between trials. This analysis excluded flecainide and propafenone, but otherwise had little effect upon the parameter estimates.

Proarrhythmic events

In total, 13 trials were included9,18,21,23,2628,32,37,42,46,48,53 in which 7860 patients were randomized and 501 patients experienced at least one proarrhythmic event as per Friedman definition13. Trials included 296 patients randomized to sotalol, 3443 patients randomized to dronedarone, 307 patients randomized to amiodarone, 90 patients randomized to flecainide, 412 patients randomized to propafenone, and 3312 patients randomized to placebo. The effect of each AAD compared with placebo on the incidence of proarrhythmic events is described in Figure 9.

Figure 9

Mixed treatment comparison analysis: effect of anti-arrhythmic drugs on incidence of proarrhythmic events, odds ratio, and 95% confidence intervals.

Dronedarone, propafenone, sotalol, and flecainide were all associated with statistically significant increased rates of proarrhythmic events. There was some evidence of heterogeneity between studies (covariance parameter 2.11, SE1.05, P = 0.046). Limiting the analysis to studies with at least 100 patients in each group resolved the limited heterogeneity, but had no qualitative impact upon the results.

Mixed treatment comparison vs. dronedarone

An advantage of the MTC approach is that the results may be expressed against a comparator of choice from within the network described. Table 4 describes the major outcomes for each comparison listed above compared with dronedarone.

View this table:
Table 4

Comparison between other active treatment strategies vs. dronedarone from the mixed treatment comparison: major outcomes, odds ratio and 95% confidence interval

AAD vs. DronedaroneaAll-cause mortalityAF recurrenceStrokeTreatment withdrawalsTreatment withdrawals due to AESerious adverse eventsProarrhythmia
Amiodarone2.52 (0.72–8.90)0.41 (0.29–0.57)1.29 (0.69–2.41)1.00 (0.63–1.58)1.71 (0.97–3.01)2.53 (1.02–6.28)3.75 (0.49–28.96)
Sotalol3.99 (1.16–13.82)0.75 (0.52–1.08)1.15 (0.56–2.39)0.82 (0.50–133)1.00 (0.63–1.58)1.35 (0.71–2.55)4.43 (0.69–28.42)
Flecainide1.97 (0.06–68.64)0.57 (0.33–0.99)NA1.10 (0.51–2.38)0.97 (0.29–3.23)2.12 (0.31–14.71)4.65 (0.57–38.21)
Propafenone0.65 (0.07–6.26)0.68 (0.46–1.00)NA1.11 (0.66–1.85)0.96 (0.53–1.75)1.64 (0.50–5.32)2.79 (0.75–10.44)
  • aOdds ratios are reporting amiodarone vs. dronedarone; sotalol vs. dronedarone; flecainide vs. dronedarone; and propafenone vs. dronedarone.


This MTC assessed the relative efficacy and tolerability of the main AADs used for the treatment of AF/AFl. The MTC method developed further the results from the conventional direct meta-analysis and the results of both methods remained internally consistent. The major findings from our analyses include a reduced risk of stroke with dronedarone, but not with other drugs studied, and some evidence towards increased mortality for sotalol and amiodarone compared with control/placebo therapy, that were most evident when studies with <100 subjects or with no events in any group were excluded from analysis.

Many of these results are also consistent with the systematic Cochrane review published previously8 despite differences in the inclusion/exclusion criteria used and the inclusion of new data on dronedarone. The eight trials that were included in the Cochrane review but not in the present analyses considered comparisons not involving amiodarone, dronedarone, propafenone, or flecainide. Compared with the data set investigated by the previous Cochrane review, we included six studies published after the Cochrane cut-off date and eight studies with a duration shorter than 6 months.

Risk of death

In the present analysis, results indicating an increased risk of death associated with sotalol were statistically significant and consistent in sensitivity analysis. The previous Cochrane systematic review8 also reported a trend towards increased mortality with sotalol compared with placebo, which became significant in some of the sensitivity analysis only.

Results from the current review also raise the possibility of an increase in mortality associated with amiodarone treatment compared with placebo. This result is less robust than that obtained for sotalol as it reached modest statistical significance only in the exploratory analysis which was restricted to inclusion of larger studies. A recent meta-analysis by Piccini et al.,58 which compared dronedarone and amiodarone using indirect comparison methods, found results very similar to ours, including also a non-significant trend towards greater all-cause mortality with amiodarone, despite minor differences in the inclusion criteria between the two reviews.

The Cochrane review8 and a different meta-analysis by Doyle et al.,59 found no evidence of an effect of amiodarone on mortality. In the analysis by Doyle et al., amiodarone was not shown to increase mortality (OR 0.95, 95% CI 0.81–1.16). However, the RCT by Roy et al.,60 which was not conducted in an AF population accounted alone for 72% of the weight in the analysis. In addition, the data that they used for the AFFIRM substudy are different from those we extracted from the publication. In indications other than AF (post-myocardial infarction and heart failure), the RCTs6163 also found no evidence of an effect of amiodarone on mortality.

In contrast to sotalol and amiodarone, there is robust evidence to suggest that treatment with dronedarone is not associated with an increase in mortality in patients with AF. This finding is consistent with the finding that dronedarone was associated with the lowest odds of proarrhythmic events. The number of patients randomized to dronedarone and compared with placebo is the largest of all AADs studied. Dronedarone may carry an increased risk of mortality in the specific population of patients with severe, decompensated heart failure, as observed in the ANDROMEDA trial.64 However, in that study, the inclusion criteria were symptomatic heart failure and not AF.

Stroke and other outcomes

Our analyses indicated that dronedarone may be associated with a reduced risk of stroke, compared with placebo. The estimates for amiodarone and sotalol for this outcome are similar (OR 0.89 and 0.80), but have wide CIs and are not significant. The analysis of this outcome is limited because this endpoint was reported in few studies, especially in larger and more recent studies, so it cannot be excluded that others AAD actually had a significant effect in this outcome. In addition, antithrombotic therapies may not have been held constant between arms in all studies. Results for dronedarone are very encouraging but cannot be considered conclusive.

All of the reviewed AADs appear better than placebo at maintaining SR but all are associated with a greater likelihood than placebo of discontinuation due to AEs. Amiodarone, compared with placebo, was associated with the largest treatment effect in reducing AF recurrence but also with the highest rate of withdrawal due to AEs. Moreover, amiodarone was associated with the highest rate of patients experiencing at least one SAE, although this estimate was not statistically significant.

Practical value of differences between drugs

The ATHENA trial9 had an overall mortality rate in the placebo group of 6%. Applying the results of the MTC to this mortality rate implies a number needed to treat for an additional death of 16 (95% CI 4 to ∞) for amiodarone, and 8 (95% CI 3–891) for Sotalol.


Conventional direct meta-analyses include only those trials which provide evidence for a specific comparison, and thus can provide inconsistent results. For example, in the reporting of the direct meta-analyses results, we note that dronedarone treatment significantly increased treatment withdrawals for any reason compared with amiodarone [OR 1.69 (95% CI 1.16–2.44)]. Yet in the next paragraph when presenting only withdrawals secondary to AEs, we report that the highest OR was reported for amiodarone compared with placebo where the odds of withdrawal were estimated at 8.41 (95% CI 2.17–32.63). Such limitations of direct meta-analysis are addressed by the MTC, which finds that there were no differences between agents in overall withdrawal compared with placebo, but significantly increased withdrawal secondary to AEs with amiodarone, sotalol, and dronedarone compared with placebo.

Thus the MTC approach may be argued to make optimal use of the available data in describing the relative effectiveness of different treatments. However, estimates derived indirectly from a MTC are not as strong as those derived from a directly randomized trial. The MTC method we used has the advantage that it preserves the integrity of randomization in the original trials, without resorting to the rather strong assumptions necessary in Bayesian analyses using numerical simulation techniques. Estimates are calculated using pseudo-likelihood directly from the data, rather than simulated from a distribution which is open to misspecification. Differences between trials are incorporated in the variances through the R side random effects, appropriately accounting for extra binomial variability (clustering at the trial level) while preserving the trial level effect estimates. However, supportive analysis conducted (not shown) using numerical simulation approaches provide similar answers to those presented here.

An important limitation of this, and any similar meta-analysis of AF trials, is that some of the outcomes we examined were secondary endpoints and may not have been ascertained and adjudicated in a consistent and complete manner. Historically most AF studies were designed with a focus on measuring rhythm-related endpoints (such as AF recurrence rates or time to first recurrence) and may have examined other endpoints, including the ones we reviewed, with less rigour. In addition, some of these outcomes (e.g. stroke and mortality) were rare in most of these trials. As a result, the findings in some cases are sensitive to meta-analytic methodology, and this may reflect some degree of uncertainty. In a small number of trials which randomized to different doses (Appendix Table A1), the doses groups were pooled which may introduce bias if some dosages were suboptimal.

The trials included in this review were conducted over a period of time which included substantial changes in patient management which could affect estimates derived in individual studies. Additionally, the standards for clinical trials have changed over time towards larger trials with longer follow-up. For example, the evidence for propafenone and flecainide was based upon few, smaller trials with fewer events compared with the other AADs. In addition these agents are not recommended for use in patients at raised cardiovascular risk and so the results are not as generalizable as those for the other agents.


All AADs included in this mixed treatment analysis are efficacious in delaying recurrence of AF. Amiodarone has been demonstrated to be the most effective drug in maintaining SR in many head to head comparison trials. However, there may also be differences in other outcomes between AADs. We found an increase in mortality associated with the use of sotalol and possibly with amiodarone. The use of dronedarone was associated with less serious AEs and less proarrhythmic events than the other anti-arrhythmics. The MTC provides the best available evidence on comparative efficacy and safety of these agents. Further large-scale comparative randomized morbidity, mortality, and patient-reported-outcome trials would be necessary to refine these estimates, particularly for older agents which have more limited evidence base.

Conflict of interest: N.F. received funding for research and consulting from sanofi-aventis and several other companies who manufacture products for cardiovascular diseases. L.E. is an employee of sanofi-aventis. S.M. received funding for research and consulting from sanofi-aventis. C.L.L. and M.R. received funding for consulting from sanofi-aventis.


The study was funded by sanofi-aventis.


Michelle Orme provided statistical support.


Sample SAS Code for estimating MTC models

proc glimmix data = work.stroke;

Title stroke by Drug;

class study;

model R/N = study Dronedarone Amiodarone Sotalol/dist = B solution CL;

NLOPTIONS tech = nrridg MAXITER = 5000;



StudyYearExperimental treatmentTreatment detailsCoded outcomes reportedTotal patientsNumber of patients discontinued prematurelyStudy durationQuality of randomizationLevel of blindingType of AF
AFFIRM substudy202003AmiodaroneAmiodarone, 200 mg/daya;b;d;f15420Up to 5 yearsInadequateOpen-labelNon-paroxysmal AF
SotalolSotalol, 240 mg/day13521
Aliot (Flecainide AF French Study Group)211996FlecainideFlecainide, 100–300 mg/daya;b;c;d;g481812 monthsUnclearOpen-labelParoxysmal AF (n = 44); flutter (n = 4)
PropafenonePropafenone, 600–1200 mg/day4926Paroxysmal AF (n = 45); flutter (n = 1); both (n = 3)
Bellandi (2001)362001SotalolSotalol, 120–240 mg/daya;b;d;f1061112 monthsUnclearUnclearPersistent/paroxysmal AF
PropafenonePropafenone, 450–900 mg/day1029
Benditt et al.571999SotalolSotalol, 160–320 mg/daya;b;c;d;f18413912 monthsAdequateInadequateNon-permanent AF/flutter
Boos et al.372008AmiodaroneAmiodarone 200–600 mg/daya;b;d;g17016 monthsUnclearOpen-labelPersistent AF, >1 month
Brodsky et al.651994SotalolSotalol, 80–160 mg/daya;b3924 monthsUnclearUnclearChronic/persistent AF
Carunchio et al.381995SotalolSotalol, 240 mg/daya;b;d20012 monthsUnclearOpen-labelParoxysmal AF
FlecainideFlecainide, 200 mg/day200
Channer et al.392004AmiodaroneAmiodarone, 200–800 mg/daya;b;d611112 monthsAdequateUnclearPersistent AF, >72 h
Chimienti et al. (FAPIS)401996FlecainideFlecainide, 100–150 mg b.d.a;b;d;f973012 monthsAdequateOpen-labelParoxysmal recurrent AF
PropafenonePropafenone, 150–300 t.i.d.10330
Cobbe531995PropafenonePropafenone, 600 mg/day (low-dose)b;c;g3013>12 monthsUnclearUnclearPSVT (n = 52)
ControlPlacebo low-dose22
PropafenonePropafenone, 900 mg/day (high-dose)2511PAF (n = 48) 
ControlPlacebo high-dose18
Connolly and Hoffert221989PropafenonePropafenone, 1200 mg/daya;c187Four 30-day treatment periodsUnclearAdequateParoxysmal AF
Davy et al. (ERATO)232008DronedaroneDronedarone, 400 mg b.d.a;b;c;f;g85176 monthsUnclearAdequatePermanent AF
DIONYSOS182009DronedaroneDronedarone 400 mg b.i.d.a;b;c;d;e;f;g
249967 monthsUnclearUnclearDocumented AF >72 h    
AmiodaroneAmiodarone 200–600 mg/day25569
Dogan et al.242004PropafenonePropafenone, 150 mg t.i.d.a;b;d;f58415 monthsUnclearUnclearPersistent (n = 14); Recent onset (n = 37) AF
ControlControl521Persistent (n = 15); Recent onset (n = 33) AF
Fetsch et al. (PAFAC)252004SotalolSotalol, 320 mg/daya;b;c;d;f38325512 monthsAdequateUnclearPersistent AF (>7 days)
Quinidine + VerapamilQuinidine, 160 mg + Verapamil, 80 mg bid377244
Galperin et al. (GEFACA)412001AmiodaroneAmiodarone, 200 mg/daya;b;d;f35116 monthsUnclearUnclearPermanent AF
Hohnloser et al. (ATHENA)92009DronedaroneDronedarone, 400 mg b.d.a;b;e;f;g2301696Mean follow-up 21 months (1–2.5 years)UnclearUnclearParoxysmal/persistent AF/flutter
Kochiadakis et al.421998AmiodaroneAmiodarone 200 mg/dayb;c;d;f;g35212 monthsUnclearUnclearChronic or paroxysmal AF
SotalolSotalol 360 mg/day352
Kochiadakis et al.432000AmiodaroneAmiodarone, 200 mg/daya;b;c;d;f6515≤2 yearsUnclear Unclear Paroxysmal(n = 42); Chronic (n = 23)
SotalolSotalol, 160–480 mg/day613Paroxysmal (n = 39); Chronic (n = 22)
ControlControl600Paroxysmal(n = 40); Chronic (n = 20)
Kochiadakis et al.442004AmiodaroneAmiodarone, 200 mg/daya;b;c;d;f7217∼24 monthsAdequateUnclearPersistent (n = 29); Paroxysmal (n = 43)
PropafenonePropafenone, 150 mg t.i.d.744Persistent (n = 25); Paroxysmal (n = 49)
Kochiadakis et al.452004bSotalolSotalol, 160–480 mg/daya;b;c;d;f855<48 monthsUnclearUnclearPersistent (n = 35); Paroxysmal (n = 50)
PropafenonePropafenone, 150 mg t.i.d.865Persistent (n = 34); Paroxysmal (n = 52)
ControlControl830Persistent (n = 34); Paroxysmal (n = 49)
Lau et al.541992FlecainideFlecainide 200 mg/dayb1936 monthsUnclearUnclearParoxysmal AF
QuinidineQuinidine 1200 gm/day
Lee et al. 1997461997PropafenonePropafenoneb;c;d;f;g4123 monthsUnclearUnclearSymptomatic, paroxysmal AF
Lombardi et al. (A-COMET-II)262006SotalolSotalol, 160 mg b.d.a;b;c;f;g2231966 monthsAdequateAdequatePersistent AF (>48 h, < 6 months)
AzimilideAzimilide, 125 mg b.d.211179
Manios et al.552003DiltiazemDiltiazemb3533 monthsAdequateOpen-labelPersistent AF
ControlNo antiarrhythmic370
Massacci et al.471992AmiodaroneAmiodarone 200–400 mg/dayb;d18618 monthsUnclearOpen-labelParoxysmal AF
FlecainideFlecainide 200–300 mg/day244
Meinertz et al. (ERAFT)272002PropafenonePropafenone, SR, 325 or 425 b.d.a;b;c;d;f;g200293 monthsUnclearUnclearParoxysmal AF
Pietersen and Hellemann (Danish-Norwegian Flecainide Multicenter Study Group)281991FlecainideFlecainide, 150 mg b.d.a;b;c;f;g4323 monthsUnclearUnclearParoxysmal AF/flutter
Patten et al. (SOPAT)292004SotalolSotalol, 160 mg b.d.a;b;c;d;f2645312 monthsUnclearUnclearParoxysmal AF (documented in previous month)
Quinidine + VerapamilQuinidine + verapamil (high dose)26368
Quinidine + VerapamilQuinidine + verapamil (low dose)25560
Pritchett et al.481991PropafenonePropafenone 300 mg b.d., t.i.d., or o.d.b;d;g23114 months with extension phaseUnclearOpen-labelParoxysmal supraventricular tachycardia or paroxysmal AF
Pritchett et al. (RAFT)492003PropafenonePropafenone SR, 225–425 mg b.d.a;b;d;f39769≤39 weeksUnclearAdequateSymptomatic AF
Reimold et al.301993SotalolSotalol, ≤480 mg b.d.a;b;d;f50612 monthsInadequateOpen-labelParoxysmal (n = 22); Chronic (n = 28)
PropafenonePropafenone, ≤300 mg t.i.d.504Paroxysmal (n = 25); Chronic (n = 25)
Roy et al. (The Canadian Trial of Atrial Fibrillation)502000AmiodaroneAmiodarone ∼200 mg/daya;b;c;d;e2016836 monthsUnclearOpen-labelPersistent AF
Sotalol or propafenoneSotalol or propafenone20293
Singh et al. (SAFE-T)312003AmiodaroneAmiodarone, 200–800 mg/daya;b;d;e;g2674212 monthsAdequateUnclearPersistent AF
SotalolSotalol, 160–320 mg/day26139
Singh et al.511991SotalolSotalol, 80–320 mg/daya;b;c;d;f24126 monthsUnclearUnclearPersistent (chronic) AF (2–52 weeks)
Singh et al. (EURIDIS ADONIS)322007DronedaroneDronedarone, 400 mg b.d.a;b;d;e;f82814812 monthsAdequateUnclearNon-permanent AF
Stroobandt et al.331997PropafenonePropafenone, 150 mg t.i.d.a;b;c;d;f101186 monthsUnclearUnclearChronic (n = 51); Recent onset (n = 49) AF
ControlControl354Chronic (n = 60); Recent onset (n = 40) AF
Touboul et al. (DAFNE)342003DronedaroneDronedarone, 400, 600, 800 b.d.a;b;c;d;f204226 monthsUnclearUnclearPersistent AF
de Simone et al. (VEPARAF)562003AmiodaroneAmiodaroneb;d8243 monthsUnclearUnclearPersistent AF
Amiodarone + VerapamilAmiodarone + Verapamil819
Flecainide + VerapamilFlecainide + Verapamil819
Van Gelder et al.521989FlecainideFlecainide, 100–150 mg b.d.a;b;c;d;f36512 monthsUnclearUnclearChronic AF/flutter
Vijayalakshmi et al.352006AmiodaroneAmiodarone, 200–600 mg/daya;b;d;f2716 monthsAdequateOpen-labelAF, <1 year duration
SotalolSotalol, 160–320 mg/day364
  • a, mortality; b, treatment discontinuation; c, SAE; d, AF recurrence; e, stroke; f, treatment discontinuation due to AE; g, proarrhythmia.

  • Table A1

    Presentation of the studies analysed


    View Abstract