Background: At the time of this trial, the recommended initial rhythm control strategy for symptomatic atrial fibrillation (AF) was anti-arrhythmic drugs with catheter ablation reserved for patients in whom drug therapy had already failed or was not well tolerated. This sequencing was never grounded in trials designed to test it directly. Trials like ThermoCool AF enrolled patients who failed to respond to at least one antiarrhythmic drug or AV nodal blocker category. Prior attempts to test ablation as a first line therapy (MANTRA-PAF, RAAFT-2) had been limited by intermittent rhythm monitoring that likely underdetected recurrence, high rates of crossover that diluted the intention-to-treat comparison, and results that were either inconclusive, showed only minor differences, or were underpowered. EARLY-AF was designed to overcome these specific limitations: a first-line comparison of cryoballoon ablation versus antiarrhythmic drugs and assessing AF burden using continuous implantable rhythm monitoring rather than intermittent surveillance.

Patients: Eligible patients were adults with symptomatic atrial fibrillation, previously untreated with rhythm-control drugs, who had at least one AF episode documented by ECG within the 24 months before randomization.

Patients were excluded if they had a history of regular, daily use of a class I or class III antiarrhythmic drug at therapeutic doses (some limited or remote prior exposure was permitted but true antiarrhythmic drug failure was an exclusion).

The trial enrolled 303 patients across 18 Canadian centers, randomized 1:1 to cryoballoon ablation (154) or antiarrhythmic drug therapy (149). Notably, the authors acknowledge they did not keep screening logs, so the size and characteristics of the broader pool from which these 303 patients were drawn — and therefore how selected this cohort is relative to the general treatment-naïve AF population — cannot be assessed.

Baseline Characteristics: The trial enrolled 303 patients across 18 Canadian centers, randomized 1:1 to cryoballoon ablation (154) or antiarrhythmic drug therapy (149).

The mean age of patients was 58 years and 70% were men. The population had a low burden of structural heart disease and comorbidity: mean left ventricular ejection fraction was approximately 60%, mean left atrial diameter 39 mm, and CHA₂DS₂-VASc scores 1.9. Approximately 95% had paroxysmal AF with a median disease duration of just 1 year since diagnosis and a median of 3 symptomatic episodes per month.

Procedures: The trial was open label. Patients were randomized in a 1:1 ratio to ablation or anti-arrhythmic drugs.

In the ablation arm, patients underwent pulmonary vein isolation using a 23-mm or 28-mm cryoballoon, with the procedural endpoint defined as bidirectional conduction block of all pulmonary veins confirmed after a 20-minute observation period; reconnected veins were re-ablated until block was achieved. Patients in the drug arm were started on a rhythm-control agent chosen by local practice. Medications were titrated over 3 months to the maximum tolerated dose. Patients could be switched to a second or third agent for inefficacy or intolerance.

All patients received an implantable cardiac monitor (Reveal LINQ) within 24 hours of treatment initiation, providing continuous beat-to-beat rhythm surveillance and quantification of AF burden. The authors applied a 90-day blanking period to both arms. During this period, arrhythmia recurrences did not count toward the primary endpoint.

Crossover was tightly and asymmetrically controlled. Patients could cross from drug therapy to ablation only after independent adjudication confirmed three criteria: a primary endpoint event had occurred after blanking, the recurrence was clinically severe enough to warrant a change in therapy, and it occurred despite a therapeutic drug dose. Crossover from ablation to drugs required only the first two criteria.

Endpoints: The primary endpoint was time to first recurrence of any atrial tachyarrhythmia (AF, atrial flutter, or atrial tachycardia) lasting 30 seconds or longer, occurring between 91 and 365 days after treatment initiation. Secondary endpoints included symptomatic arrhythmia recurrence, AF burden (percentage of time in AF), quality of life via AFEQT and EQ-5D, health care utilization, and safety.

Analysis was performed on the intention-to-treat principle. The sample size calculation assumed 88 events would provide 90% power to detect a 20-percentage-point difference in recurrence, with enrollment inflated to account for anticipated crossover. All endpoints were adjudicated by a committee blinded to treatment assignment.

Results: No patient crossed over from their assigned treatment strategy before the occurrence of a primary end-point event. Complete isolation was confirmed in all 152 patients who underwent the procedure. One repeat ablation was performed during the blanking period and was considered a primary endpoint event. Flecainide was the most frequently used antiarrhythmic agent, with a median daily dose of 200 mg. Overall, 103 of 149 patients (69.1%) were treated with only one antiarrhythmic drug.

At 1 year, a documented recurrence of atrial tachyarrhythmia was significantly less common in the ablation arm (42.9% vs 67.8%, HR 0.48, 95% CI: 0.35 - 0.66; p< 0.001). Symptomatic atrial fibrillation was also less frequent with ablation (11.0% vs 26.2%, HR: 0.39, 95% CI: 0.22 - 0.68). Among patients assigned to catheter ablation, the median proportion of time spent in atrial fibrillation was 0% (interquartile range, 0–0.08), compared with 0.13% (interquartile range, 0–1.60) among those assigned to antiarrhythmic drug therapy. Mean AF burden was 0.6±3.3 with ablation vs 3.9±12.4 with drug therapy.

Figure below taken from the main publication.

Quality-of-life improvements favored ablation but were more modest than the arrhythmia endpoints might suggest. AFEQT scores improved by a least-squares mean of 26.9 points with ablation versus 22.9 with drugs at 12 months, a between group difference of roughly 4 points that, while statistically directionally favorable, is of uncertain clinical significance given typical AFEQT minimal important difference thresholds. EQ-5D and EQ-VAS differences were similarly small. The proportion of patients reporting no symptoms was higher with ablation at both 6 months (83.8% vs 60.4%) and 12 months (85.1% vs 73.2%). Hospitalization lasting more than 24 hours was numerically lower with ablation (3.2% vs 8.7%).

Serious adverse events occurred in 3.2% of ablation patients versus 4.0% of drug therapy patients. The specific complication profiles differed in character: ablation was associated with three cases of phrenic nerve palsy (all resolved within one month) and no procedural deaths or thromboembolic events, while the drug arm saw adverse drug reactions including QT prolongation, presyncope, and one case of heart failure exacerbation, along with cardiac tamponade in one patient who had crossed over to ablation after drug failure.

Conclusions: In patients with atrial fibrillation, cryoballon ablation reduced AF recurrence from day 91 - 365 compared to anti-arrhythmic drugs. The primary endpoint [time to first recurrence of any atrial tachyarrhythmia (AF, atrial flutter, or atrial tachycardia) lasting 30 seconds or longer, occurring between 91 and 365 days] occurred in 42.9% of the patients in the ablation arm vs 67.8% in the drug arm corresponding to a number needed to treat of approximately 4.

While the treatment effect is very large, several points to keep in mind:

• The lack of screening logs is an important limitation. Without information on the number of patients assessed before enrollment of the 303 participants, it is difficult to determine how representative the study cohort is of the broader treatment-naïve atrial fibrillation population in routine clinical practice.

• The trial was explicitly underpowered for cardiovascular outcomes. Therefore, the trial can speak only to rhythm control, symptom burden, and quality of life, not to whether earlier ablation reduces stroke, heart failure hospitalization, or mortality relative to drug therapy.

• The one-year follow-up is not long enough to draw conclusions about long-term durability of rhythm control or disease progression.

• Pulsed-field ablation is becoming increasingly popular, and since this trial used only cryoballoon ablation, the findings may not fully apply to patients treated with newer ablation techniques, which some argue are superior (this needs to be confirm in a large trial).

• Most importantly, the 3-month blanking period accounts for 25% of the total trial follow-up and leaves an important gap in our understanding of what happened during this period. Although AF recurrence during the blanking period does not necessarily indicate treatment failure, reporting these episodes is still important. How do we know that ablation does not increase arrhythmia burden during this time? Furthermore, the overall AF burden was low in both groups, with a median of 0% (interquartile range, 0–0.08) after ablation versus 0.13% (interquartile range, 0–1.60) with antiarrhythmic drug therapy. The mean AF burden was also lower with ablation (0.6±3.3% vs. 3.9±12.4%), an absolute difference of 3.3 percentage points. Given the large absolute reduction in the primary endpoint of 24.9%, these findings suggest that most AF episodes were brief and may not have been clinically significant.

Background: Prior trials showed that patients who undergo ablation are more likely, compared to anti-arrhythmic drugs, to be in is sinus rhythm over long follow up. Whether restoring sinus rhythm through ablation actually reduced mortality, stroke, or other major adverse events remained unknown, despite ablation’s use having already expanded into higher-risk populations without such outcome data to support it. Prior to CABANA no large trial had tested whether ablation improves outcomes compared to medical therapy in a broad, representative atrial fibrillation (AF) population. CABANA sought to answer that question - an investigator-initiated, NIH- and industry-funded, multicenter, open-label trial designed to determine whether a strategy of catheter ablation reduces hard outcomes compared with guideline-directed drug therapy.

Patients: Eligible patients were 65 years or older, or younger than 65 with at least one stroke risk factor (hypertension, heart failure, prior stroke, diabetes, or other heart disease), and had either two or more episodes of paroxysmal AF or one episode of persistent AF in the preceding six months. Patients were excluded if they had undergone prior left atrial catheter ablation, had already failed two or more antiarrhythmic drugs, had reversible causes of atrial fibrillation plus many others (see table 1 in Supplement 3 of the published article).

Notably, sites were not required to maintain screening logs and therefore, the total pool of patients assessed for eligibility, and how selected the enrolled cohort actually was, cannot be determined.

Baseline Characteristics: A total of 2,204 patients were randomized across 126 centers in 10 countries: 1,108 to catheter ablation and 1,096 to drug therapy. Median age was 68 years, just over a third of patients were women, and roughly 10% belonged to racial or ethnic minorities in each arm. Approximately 81% had hypertension, 26% diabetes, 19% coronary artery disease, 15% a history of heart failure, and 10% a prior stroke or TIA, with a median CHA₂DS₂-VASc score of 3 and 57% scoring above 2. AF type was roughly 43% paroxysmal, 47% persistent and 10% long-standing persistent. Baseline characteristics were well balanced between arms.

Procedures: The trial was open-label. Patients were randomized in a 1:1 ratio to catheter ablation or medical therapy.

Ablation patients underwent pulmonary vein isolation, with additional ancillary lesion sets (linear, ganglion plexus, electrogram-based) left to investigator discretion; operators were required to have performed at least 100 prior cases. Drug therapy patients received guideline-directed rate or rhythm control medications (rate control was recommended as a first line). Anticoagulation guided by contemporaneous guidelines in both arms.

Enrollment proved slower than anticipated, and event rates ran well below projections; in response, the Data Safety and Monitoring Board recommended a substantial protocol restructuring in 2013, roughly midway through enrollment. The originally planned all-cause mortality primary endpoint was demoted to key secondary status, a new composite (death, disabling stroke, serious bleeding, or cardiac arrest) was elevated to primary, and the sample size target was reduced from 3,000 to 2,200 patients. While this restructuring was performed blinded to treatment-specific outcomes and is statistically defensible, mid-trial endpoint switching of this magnitude invites legitimate scrutiny about how much the final analysis reflects the question originally posed versus the question that turned out to be answerable with the patients actually accrued.

Treatment adherence was also considerably low. Among patients randomized to ablation, only 90.8% (1,006 of 1,108) actually underwent the procedure, with the remainder declining largely due to patient or family refusal. Among those randomized to drug therapy, 27.5% (301 of 1,096) ultimately crossed over to ablation during follow-up. This degree of bidirectional crossover meaningfully dilutes the intention-to-treat comparison and is a central thread running through the trial’s interpretation.

Endpoints: The primary endpoint, following the 2013 amendment, was a composite of death, disabling stroke, serious bleeding, or cardiac arrest, adjudicated blindly by an independent clinical events committee. All-cause mortality, death or cardiovascular hospitalization, and AF recurrence were reported as three of thirteen prespecified secondary endpoints. AF recurrence was counted after 3-month blanking period in both treatment arms.

The trial’s original design assumed a 12% three-year mortality rate in the drug arm and a 30% relative risk reduction with ablation; both assumptions would prove substantially too optimistic, with major consequences for the trial’s ability to answer its own question.

The main analysis was performed based on the intention-to-treat (ITT) principle. Prespecified sensitivity analyses - treatment-received (ablation modeled as a time-dependent covariate) and per-protocol (censoring drug-arm crossovers at time of ablation) - were planned in advance specifically to address anticipated crossover, alongside a formal interim analysis using O’Brien-Fleming boundaries.

Results: The median follow-up time was 48.5 months. In the intention-to-treat analysis, the primary composite endpoint occurred in 8.0% of the ablation group versus 9.2% of the drug therapy group (HR 0.86, 95% CI 0.65-1.15, P=.30). Four-year Kaplan-Meier event rates were 7.2% versus 8.9% (absolute difference 1.7%). When breaking down the individual components of the primary endpoint - none reached statistical significance (rate with ablation mentioned first): Death (5.2% vs 6.1%, HR 0.85, 95% CI 0.60-1.21), disabling stroke (0.3% vs 0.6%, HR 0.42, 95% CI 0.11-1.62), serious bleeding (3.2% vs 3.3%, HR 0.98, 95% CI 0.62-1.56) and cardiac arrest (0.6% vs 1.0%, HR 0.62, 95% CI 0.24-1.61).

The one ITT result that did reach conventional significance was the secondary endpoint of death or cardiovascular hospitalization (51.7% vs 58.1%, HR 0.83, 95% CI 0.74-0.93, P= 0.001). This endpoint warrants particular caution, however, since hospitalization was adjudicated by unblinded site investigators in an open-label trial introducing bias.

AF recurrence (counted after 3-month blanking period in both treatment arms), assessed in the 1,240 patients using the trial’s ECG recording system, favored ablation substantially (49.9% vs 69.5% at three years, adjusted HR 0.52, 95% CI 0.45-0.60, P<.001). Notably, 215/1006 (21.4%) of patients who received ablation in the ablation arm, required at least one repeat procedure (25 of these patients had their repeat ablation during the 3-month blanking period).

The prespecified subgroup analysis found no significant interaction across age, sex, AF type, heart failure status, or CHA₂DS₂-VASc risk category, although there was a trend for better outcomes with ablation in younger patients (<65).

The sensitivity analyses produced the most provocative numbers in the trial. The treatment-received analysis yielded an HR of 0.67 (95% CI 0.50-0.89, P=0.006) for the primary endpoint and 0.60 (95% CI 0.42-0.86, P=0.005) for all-cause mortality. Per-protocol analyses using 6- and 12-month ablation windows produced HRs of 0.74 (95% CI, 0.54-1.01) and 0.73 (95% CI, 0.54-0.99) for the primary endpoint, and 0.69 (95% CI, 0.47-1.01) and 0.68 (95% CI, 0.47-0.99) for mortality, respectively - all more favorable to ablation than the intention-to-treat estimates.

Adverse events with ablation were infrequent when performed by experienced operators, with cardiac tamponade (0.8%) the most common serious complication and no atrial-esophageal fistula observed; the drug therapy arm saw thyroid disorders (1.6%) and pro-arrhythmia (0.8%) as the most notable adverse events.

Conclusions: In patients with atrial fibrillation, ablation was not superior to drug therapy for the primary outcome of death, disabling stroke, serious bleeding, or cardiac arrest, over a median of 48.5 months of follow up.

This is the largest trial to date that examined ablation in patients with atrial fibrillation and the investigators should be commended for conducting it.

Several aspects are worth highlighting.

• First, sites were not required to maintain screening logs, so the total pool of patients assessed for eligibility, and how selective the enrolled cohort actually was cannot be determined.

• Second, cardiovascular hospitalization was adjudicated by site investigators, putting it at higher risk of bias given the trial’s open-label design.

• Third, the ITT analysis should be regarded as the primary analysis and the one carrying the lowest risk of bias. The treatment-received and per-protocol analyses both introduce selection bias, and the authors are appropriately candid about this. These approaches discard randomization’s protection against selection bias, since patients who decline assigned ablation, or who seek out ablation after being randomized to drugs, are self-selected in ways that plausibly correlate with prognosis independent of any true treatment effect.

• Fourth, substantial bidirectional crossover further complicates matters - 9.2% of patients assigned to ablation never received it, while 27.5% of patients assigned to drug therapy crossed over to ablation. This diluted the ITT comparison in ways that the treatment-received and per-protocol analyses only partially, and imperfectly, correct for.

Background: Prior trials such as AF-CHF and DIAMOND-CHF showed that drug rhythm control does not reduce mortality in patient with atrial fibrillation and heart failure. By the time CASTLE-AF was conceived, the ablation-versus-drug studies had shown that ablation improved rhythm control, left ventricular ejection fraction, exercise capacity, and quality of life relative to antiarrhythmic drugs (studies had limitations as we previously discussed). However, no trial had been adequately powered to test whether any of that translated into the hard outcomes such as death or hospitalization. CASTLE-AF was designed explicitly to separate these possibilities by testing catheter ablation, rather than a drug, against a hard composite endpoint of death or heart failure hospitalization. The trial was industry-sponsored by Biotronik, whose device was mandated for study entry, and the sponsor assisted with data management and statistical quality control; the manuscript states explicitly that Biotronik had no role in trial design or execution, and multiple authors disclose extensive consulting and research relationships with device manufacturers, most heavily Biotronik itself.

Patients: Eligible patients had heart failure with a history of symptomatic paroxysmal or persistent atrial fibrillation (AF), no response to (or unacceptable side effects from, or unwillingness to take) antiarrhythmic drugs, NYHA class II–IV, left ventricular eejction fraction ≤35%, and a mandatory implanted device, specifically a Biotronik-manufactured ICD or CRT-D with automatic remote-monitoring capability. Major exclusions were candidacy for heart transplantation or a planned cardiovascular intervention; the supplementary appendix reveals a substantially longer exclusion list, including prior left atrial ablation, left atrial diameter >6 cm, life expectancy ≤12 months, uncontrolled hypertension, and untreated thyroid disease. The device-manufacturer specificity is the most distinctive eligibility feature of this trial relative to its predecessors: it was operationally necessary to standardize remote monitoring, but it also means eligibility was constrained not just by clinical characteristics but by which ICD vendor a patient happened to already have - a criterion of pure trial-logistics origin that nonetheless shaped the entire funnel from screening to enrollment, as the supplementary screening breakdown makes clear (975 of 2615 screen failures, or 32.4% of all patients screened, were excluded specifically because their implanted device was not Biotronik-compatible, more than double any other single exclusion reason).

Baseline Characteristics: From January 2008 through January 2016, 3013 patients were assessed across 33 sites in Europe, Australia, and the United States, with just 398 enrolled - a screening-to-enrollment ratio (13.2%) that reflects both the device-vendor constraint and the trial’s demanding combination of heart failure severity and AF history requirements. After a 5-week run-in period used to optimize guideline-directed heart failure therapy, 363 patients remained for the primary analysis: 179 assigned to ablation and 184 to medical therapy.

Median age was 64, and 86% were men. Median left ventricular ejection fraction was 32%, and left atrial diameter was 49 mm. A few meaningful baseline imbalance stands out: cause of heart failure differed significantly between arms (ischemic etiology 40% in the ablation group vs. 52% in the medical-therapy group, P=0.022). The supplementary appendix shows this imbalance traces back to enrollment rather than differential run-in attrition, and while the trial’s conclusions do not appear to hinge on it, an over-representation of nonischemic (generally more reversible) cardiomyopathy in the ablation arm is worth keeping in mind when interpreting the results. Furthermore, in the randomized population, less patients in the ablation arm had diabetes or were taking digitalis (28% vs 37% and 20% vs 30% respectively).

History of amiodarone failure was documented in 46% of patients overall, meaning the comparator arm was not amiodarone-naive rhythm control but a mixed, largely rate-control-oriented “medical therapy” strategy (Rhythm-control strategy was used in approximately 30% of the patients in the medical-therapy arm).

Procedures: Patients were randomized in a 1:1 ratio to ablation or medical therapy. The trial was open label.

Ablation aimed for isolation of all pulmonary veins with restoration of sinus rhythm; additional lesions were left to operator discretion, and all operators were required to have performed at least 50 prior AF ablation procedures. Left atrial thrombus was excluded by transesophageal echocardiography before every procedure, and post-ablation anticoagulation with warfarin was mandated for a minimum of six months. Of the 179 patients randomized to ablation, only 151 (84.4%) actually received it; 28 (15.6%) crossed over to medical therapy, most commonly because a left atrial appendage thrombus failed to dissolve within the allotted procedural window, for other medical reasons, or by patient decline. Among those who underwent ablation, pulmonary vein isolation was achieved in 149 of 151 (98.7%), with additional lesions in just over half; the mean number of procedures per treated patient was 1.3 ± 0.5, and repeat ablation for recurrence occurred in roughly a quarter of patients at a median of over a year after the index procedure. On the medical-therapy side, 18 of 184 patients (9.8%) crossed over to ablation, and a rhythm-control strategy (antiarrhythmic drugs or ablation) was used in only about 30% of the medical-therapy group throughout follow-up, with the substantial majority managed by rate control alone (Figure S3 in the supplement). This is an important design choice to sit with: the “medical therapy” comparator in CASTLE-AF was predominantly a rate-control strategy, not a rhythm-control strategy using drugs, since the protocol deliberately avoided mandating a specific pharmacological approach given the prior AF-CHF and DIAMOND-CHF findings that drug rhythm control did not show mortality benefit.

Endpoints: The primary endpoint was the composite of death from any cause or hospitalization for worsening heart failure which was adjudicated by an independent committee blinded to treatment assignment. Secondary endpoints included the individual components of the composite, cardiovascular death, cerebrovascular accident, cardiovascular hospitalization, and all-cause hospitalization; in the ablation group, procedure-related adverse events and AF-free intervals were tracked separately. The trial used a three-stage adaptive group-sequential design, powered to detect a hazard ratio of 0.67 with 80% power at a two-sided alpha of 0.05, requiring 195 primary end-point events at final analysis. This is the single most consequential methodological feature of the trial: enrollment and event accrual were both slower than projected, and by December 2016 it became clear the target of 195 events would not be reached in a reasonable timeframe. The trial was stopped after only 133 events using a conditional rejection probability method (Müller-Schäfer) to preserve the overall alpha level. This a legitimate, pre-specifiable statistical adjustment, but one that means the trial’s primary result rests on materially fewer events than originally planned, even though the supplementary appendix notes that a post-hoc check confirmed 133 events would also have crossed the pre-specified O’Brien-Fleming boundary under the original plan. The intention-to-treat-adjacent “full analysis set” used for the primary analysis excluded 34 of 397 randomized patients who dropped out during the run-in period before effective randomization at the baseline visit - a modification from the original enrollment-based ITT principle, though the supplement shows sensitivity analyses including these patients yielded materially identical hazard ratios (table S7).

The authors wrote this in the original and it’s very important to keep in mind when interpreting the results: “The treatment groups were compared on a modified intention-to-treat basis. This analysis excluded patients who had died or were withdrawn from the trial during the run-in period. It also excluded end-point events occurring during the run-in period and included only deaths and not other events during the first 12 weeks after the baseline visit (the “blanking period” after ablation, with an identical period of event exclusion after baseline in the medical-therapy group)”.

Results: After a median follow-up of 37.8 months (mean 37.6), the primary composite endpoint occurred in 51 of 179 ablation patients (28.5%) versus 82 of 184 medical-therapy patients (44.6%); hazard ratio 0.62, 95% CI 0.43 - 0.87, P=0.007 by Cox regression and P=0.006 by log-rank test. The number needed to treat to prevent one primary endpoint event at 36 months was 8.3. All-cause mortality alone was significantly lower with ablation (13.4% vs 25.0%; HR 0.53, 95% CI 0.32–0.86, P=0.01), and Kaplan-Meier curves show this mortality benefit did not clearly emerge until after approximately three years of follow-up. Heart failure hospitalization was also lower with ablation (20.7% vs 35.9%; HR 0.56, 95% CI 0.37–0.83, P=0.004). Cardiovascular hospitalization was also reduced (HR 0.72, 95% CI 0.52–0.99, P=0.04), but all-cause hospitalization was not significantly different (63.7% vs 66.3%; HR 0.99, 95% CI 0.77–1.28, P=0.96).

Figure below taken directly from the primary publication:

The prespecified subgroup analysis showed a significant interaction between baseline left ventricular ejection fraction (LVEF) and treatment effect (P=0.01), such that patients with LVEF ≥25% derived benefit from ablation (HR 0.48, 95% CI 0.31–0.74) while those with LVEF <25% did not (HR 1.36, 95% CI 0.69–2.65). This should be interpreted with extreme caution, given that it's a subgroup analysis, and especially given the trial's small overall sample size and even smaller subgroup sizes.

LVEF improved substantially more with ablation: median absolute increase of 8.0 percentage points at 60 months versus 0.2 in the medical-therapy group (P=0.005). AF burden, extracted from device memory, was significantly lower in the ablation arm: mean AF burden fell from roughly 51% at baseline to the 20–27% range by 12 months in the ablation group and stayed there through 60 months, while the medical-therapy group’s AF burden remained essentially unchanged around 48–54% throughout follow-up and actually rose to 64% by 60 months. Sinus rhythm maintenance at 60 months was present in 63.1% of ablation patients versus 21.7% of medical-therapy patients (P<0.001).

Regarding procedural safety: three ablation patients had pericardial effusion (one requiring pericardiocentesis), three had severe bleeding requiring transfusion, and one developed pulmonary vein stenosis; no strokes or deaths were attributed to the ablation procedure itself.

Conclusions: CASTLE-AF is the first trial to demonstrate, with adequate design intent and a hard composite primary endpoint, that catheter ablation reduces death and heart failure hospitalization compared with usual medical management in patients with AF and heart failure with reduced ejection fraction. The effect size was large - the number needed to treat to prevent one primary endpoint event at 36 months was 8.3.

The trial's primary analysis was modified intention-to-treat, an approach that opens the door to selection and attrition bias. That said, the we commend the investigators for also reporting the intention-to-treat results, and the findings held up unchanged.

This is an important trial, but there are several key considerations to keep in mind when interpreting and applying these results:

• The trial compared ablation to patients with no response to (or unacceptable side effects from, or unwillingness to take) antiarrhythmic drugs. By design, then, the trial did not test ablation as first-line therapy, nor did it compare ablation against patients who can tolerate antiarrhythmic drugs and/or in whom antiarrhythmic drugs are effective.

• The heart failure hospitalization endpoint is high risk for bias given open-label design and importantly, it did not count events occurring during the 12-week post baseline visit/ ablation (see the last paragraph of the "Endpoint" section above) - we see no justification for such exclusion.

• The trial stopped early, at 133 of a planned 195 events. While statistically robust under the pre-specified adaptive rules, this result rests on a narrower evidence base than originally intended.

• The trial's sample size is small, and as we've discussed in reviews of other trials, effect sizes from smaller studies don't always hold up in large trials.

• The loss of follow up is large and unequal between both arms [23 patients (12.8%) in the ablation arm vs 10 (5.4%) in the medical arm]. Patients lost to follow-up are essentially unknowns: we don't know whether they died, were hospitalized, or simply felt too well to keep returning for visits. This differential attrition bias can significant modify the effect size.

• Finally, how much of the treatment effect could be explained by the imbalance of baseline characteristics? (ischemic cardiomyopathy, diabetes and digitals use were all lower in the ablation arm at randomization - 42% vs 51%, 28% vs 37% and 20% vs 30%, respectively).

Background: APAF-CRT was conceived as a single research program executed in two consecutive, overlapping phases rather than as two independent trials, and it’s worth treating it that way in appraisal rather than reviewing the two papers as separate studies. The morbidity trial (2018) tested whether AV junction ablation combined with biventricular pacing outperforms pharmacological rate control for reducing heart failure (HF) morbidity and hospitalization in permanent atrial fibrillation (AF) with narrow QRS; the mortality trial (2021) continued enrollment and extended follow-up in that same investigative program specifically to test whether the same strategy improves survival, a question the morbidity trial was explicitly not powered to answer but nonetheless showed an encouraging directional signal for. The rationale throughout rests on a specific physiological argument: AV junction ablation reliably slows and regularizes the ventricular response in permanent AF, but earlier studies of ablation with plain right ventricular (RV) pacing had shown neutral effects on HF progression, hospitalization, and mortality despite symptomatic improvement, plausibly because RV pacing itself induces left ventricular dyssynchrony resembling left bundle branch block in roughly half of patients. Biventricular pacing was hypothesized to neutralize that dyssynchrony penalty, allowing the benefit of near-perfect rate regularization to manifest without a competing harm, a mechanistic hypothesis that had support from smaller studies and registries but had never been tested in a dedicated randomized trial against pharmacological rate control specifically in a narrow-QRS population (a population that, by conventional criteria, would not otherwise qualify for cardiac resynchronization therapy “CRT”). Both phases were investigator-initiated and independent, sponsored by a non-profit Italian foundation that received an unrestricted grant from Boston Scientific for the morbidity trial and continued that funding relationship into the mortality phase; the manuscripts state Boston Scientific had no role in trial design, data collection, analysis, or the publication decision in either phase.

Patients: Both phases used essentially the same eligibility criteria: severely symptomatic permanent AF of more than six months’ duration, judged unsuitable for or already having failed AF ablation; narrow QRS (≤110 ms); and at least one hospitalization for heart failure in the prior year. This is a population deliberately positioned downstream of the ablation-for-rhythm-restoration decision - these are patients for whom pulmonary vein isolation has already been ruled out or has failed, making AV junction ablation with CRT a different branch point in the treatment algorithm entirely, not a head-to-head competitor with AATAC- or CASTLE-AF-style rhythm-restoration strategies. There is no ejection fraction floor for entry in either phase; randomization was stratified by baseline EF (≤35% vs >35%). The morbidity trial additionally excluded NYHA class IV patients with systolic blood pressure ≤80 mmHg despite optimized therapy, recent myocardial infarction, and patients with previously implanted devices - exclusions not explicitly re-stated in the mortality trial paper but presumably carried forward given the shared protocol lineage. The mortality trial’s population (133 patients: 63 ablation + CRT, 70 drug) is a superset of the morbidity trial’s population (102 patients: 50 ablation + CRT, 52 drug): recruitment continued after the morbidity trial was stopped early for efficacy in May 2018, adding roughly 31 more patients across 11 centers (one more site than the morbidity trial’s 10) through December 2020, with all patients - both the original 102 and the additional enrollees - followed for up to four years for the mortality endpoint. This continuous-enrollment design means the mortality trial is not an independent replication in a fresh sample but a substantially overlapping, extended-follow-up analysis of largely the same patients plus incremental accrual.

Baseline Characteristics: The two phases’ baseline tables are close enough in structure and values. In the morbidity trial (102 patients), the average age was 72 years, and 55% of patients were men. The average ejection fraction was 41% in both arms, with 42% of patients having an EF ≤35%. Duration of permanent AF was approximately 16 months (combined IQR roughly 8–40), and 65% of patients were NYHA class ≥III. Comorbidity burden was broadly similar: hypertension in 73%, diabetes in 21%, and coronary heart disease in 31% (the largest single difference between arms, 26% vs. 36%, though not flagged as statistically significant in the paper). Digoxin use after 30-day optimization averaged 47% overall, but was significantly lower in the ablation + CRT arm (36% vs. 58%, P=0.03) - the only baseline/post-optimization variable reaching significance in Table 1.

In the mortality trial’s larger (133 patients), longer-recruited cohort, baseline characteristics shifted only modestly from the morbidity-trial values, consistent with the additional ~31 patients resembling those already enrolled: mean age rose slightly to 73 years, sex distribution was more balanced at 54% men, and mean ejection fraction (EF) was unchanged at 41% overall, with a slightly higher proportion in the low-EF stratum (42% with EF ≤35%). Duration of permanent AF was closely matched between arms in this larger cohort (approximately 19 months, median), unlike the morbidity trial’s somewhat wider gap - suggesting the additional enrollees included relatively more longer-standing AF in the ablation + CRT arm, balancing out the earlier imbalance. NYHA class ≥III remained common, at 69% overall. Symptom burden, captured by the Specific Symptoms Scale, was closely matched at approximately 29 overall. Comorbidities were again broadly similar between arms (hypertension 74%, diabetes 24%), though coronary heart disease remained numerically higher in the drug arm in both phases (31% overall) without being highlighted as a significant imbalance. As in the morbidity trial, digoxin use after optimization was substantially and significantly higher in the drug arm (46% overall; 32% vs. 60%, P<0.001).

Overall, both phases achieved reasonably well-matched randomization.

Procedures: Patients were randomized in a 1:1 ratio to atrioventricular junction ablation and biventricular pacemaker (Ablation + CRT) or pharmacological rate control.

Procedures were essentially identical across phases. Right-sided AV junction ablation was attempted first, with a left-sided approach added if right-sided ablation failed to achieve persistent third-degree AV block; repeat ablation was recommended if AV block regressed. Any commercially available CRT-P or CRT-D device was permitted, with the RV lead in the apex and the LV lead targeted to the basal-mid free wall; no atrial lead was used given the permanent-AF indication. Device implantation and ablation were both completed within 30 days of randomization in both phases (median 2.5–4 days to CRT implant, 6 days to ablation). In the drug arm, rate-control optimization targeted a resting heart rate below 110 bpm, but this optimization was left to individual investigator discretion rather than a standardized protocol. Crossover from drug to ablation + CRT occurred in both phases and increased in magnitude with longer follow-up: 12 of 52 drug-arm patients (23%) crossed over in the morbidity trial at a median of 135 days, versus 18 of 70 (26%) in the mortality trial, the large majority in both cases triggered by reaching the HF hospitalization endpoint and permitted per protocol to cross over.

Endpoints: This is where the two phases genuinely differ and where the appraisal needs to be precise. The morbidity trial’s primary endpoint was a composite of death due to HF, hospitalization due to HF, or worsening HF. Secondary endpoints included total mortality, HF hospitalization, and worsening HF individually. The mortality trial’s primary endpoint, by contrast, was the harder and more specific outcome of time to all-cause mortality alone.

Sample size justification differed accordingly: the morbidity trial based its power calculation on historical RV-pacing data from an earlier, related trial (APAF), estimating 137 patients per arm would be needed but stopping early at interim analysis after only 25 events (roughly 50% of patients enrolled) once the prespecified stopping boundary was crossed; the mortality trial, lacking any prior mortality-specific estimate for this exact intervention, built its power calculation on the morbidity trial’s own sparse mortality data (8 total deaths, HR 0.30). The mortality trial similarly stopped early at interim analysis, after 27 of a targeted 32 events, when the Data Safety Monitoring Board judged the prespecified stopping boundary satisfied.

Analysis was performed based on the intention-to-treat principle.

Results: In the morbidity trial, over a median follow-up of 16 months, the primary composite endpoint (death due to HF, HF hospitalization, or worsening HF) occurred in 10 of 50 ablation + CRT patients (20%) versus 20 of 52 drug patients (38%) - HR 0.38, 95% CI 0.18–0.81, P=0.013. HF hospitalization specifically was significantly reduced (5 [10%] vs. 13 [25%]; HR 0.30, 95% CI 0.11 - 0.84, P= 0.024), and quality of life measures (Specific Symptom Scale score, EHRA class) improved significantly more with ablation + CRT at one year. A prespecified subgroup analysis found a significant interaction between the primary endpoint and both baseline ejection fraction and symptom severity: patients with ejection fraction ≤35% benefited more (HR 0.18, 95% CI 0.05–0.66) than those with ejection fraction >35% (HR 0.62, 95% CI 0.23–1.70, interaction P=0.05), and more symptomatic patients (Specific Symptom Scale score >31) benefited more than less symptomatic patients (interaction P=0.001) - a materially different subgroup signal than the later mortality trial would produce.

In the mortality trial, over a longer median follow-up of 29 months in the larger, continued-enrollment population, all-cause mortality occurred in 7 of 63 ablation + CRT patients (11%) versus 20 of 70 drug patients (29%) - HR 0.26, 95% CI 0.10–0.65, P=0.004). The secondary composite of death or HF hospitalization was similarly reduced (18 [29%] vs. 36 [51%]; HR 0.40, 95% CI 0.22–0.73, P=0.002). In the prespecified subgroup analysis by ejection fraction, all-cause mortality benefit was seen in patients with ejection fraction >35% (HR 0.27, 95% CI 0.08–0.84; P=0.024), with no significant interaction versus patients with ejection fraction ≤35% (HR 0.34, 95% CI 0.06–1.92; P=0.22).

Figure below from the mortality trial -

Conclusions: Taken as a single continuous research program, APAF-CRT provides a reasonably coherent and clinically important body of evidence that AV junction ablation combined with biventricular pacing improves both HF morbidity and, with longer follow-up in an extended cohort, all-cause mortality relative to pharmacological rate control in elderly, severely symptomatic, permanent-AF patients with narrow QRS and a prior HF hospitalization.

The program’s central strength is its internal consistency: the morbidity trial’s early, hypothesis-generating mortality signal was subsequently tested as a dedicated primary hypothesis in an extended cohort and reproduced statistically significant bounds. This is a reassuring pattern of a preliminary signal holding up under a more rigorous, prespecified test.

That said, several limitations compound rather than resolve across the two phases. Both trials were stopped early at interim analysis, a practice that inflates observed effect sizes; both rest on very small absolute event counts (25 and 27 events, respectively); the mortality trial’s sample size calculation was itself built on the morbidity trial’s unstable 8-event mortality estimate; and because the mortality trial’s cohort substantially overlaps with (rather than independently replicates) the morbidity trial’s population, the mortality phase functions more as an extended follow-up of largely the same patients than as a fully independent confirmatory trial.

The population itself is also narrowly defined (elderly, referred to cardiology departments rather than drawn from the general hospitalized-HF population), and explicitly restricted to patients for whom AF ablation was not an option which limits generalizability to broader, younger, or less severely affected AF-HF populations, and the pharmacological comparator arm’s non-protocolized rate-control optimization leaves some residual uncertainty about whether a more aggressive drug strategy might have narrowed the gap.

Finally, the trials had small sample size (102 and 133 patients). We encourage readers to review articles below - Smaller studies sometimes correlate with the results of larger trials, but this consistency is not guaranteed.

N Engl J Med 1997;337:536-542
JAMA 1996;276:1332-1338

Background: Atrial fibrillation (AF) and heart failure (HF) frequently coexist, and the prevalence of AF rises with worsening NYHA class, from roughly 5% in class I to approximately 50% in class IV. Prior randomized trials of pharmacological rhythm control (using antiarrhythmic drugs, AADs) had failed to show superiority over rate control in patients with or without HF, and the available AADs for maintaining sinus rhythm in HF (amiodarone and dofetilide) both carry significant toxicity. Catheter ablation had shown promise as an alternative method of achieving sinus rhythm without drug-related side effects, and the PABA-CHF trial had shown pulmonary-vein isolation to be superior to AV-node ablation with biventricular pacing in HF patients with AF, though that benefit sometimes required more than one procedure. Whether catheter ablation was superior to amiodarone specifically - the mainstay AAD used in this population - for the treatment of persistent AF in patients with HF had not been tested in a randomized trial. The AATAC (Ablation vs Amiodarone for Treatment of Atrial Fibrillation in Patients with Congestive Heart Failure and an Implanted ICD/CRT-D) trial sought to address this question.

Patients: Eligible patients were 18 years of age or older with persistent AF, NYHA class II or III heart failure, an LVEF of 40% or less within the past 6 months, and an already-implanted dual-chamber implantable cardioverter-defibrillator (ICD) or cardiac resynchronization therapy defibrillator (CRT-D).

Patients were excluded if AF had a reversible etiology, if they had valvular or coronary heart disease requiring surgical intervention, early postoperative AF (within 3 months of cardiac surgery), a life expectancy of 2 years or less, a prolonged QT interval, hypothyroidism, severe pulmonary disease, or liver failure. Patients already receiving a regular (≥200 mg/day) dose of amiodarone were also excluded, although 27 patients already on a low dose (<200 mg/day) were permitted to enroll.

Baseline characteristics: Patients were enrolled from multiple centers in the United States and Europe (Italy, France, and Czech Republic). Of 866 patients screened, 331 (38.2%) were eligible, and 203 consented and were randomized to catheter ablation (102 patients) or amiodarone (101 patients).

The average age of patients was 61 years and 74% were men. The average AF duration prior to enrollment was approximately 8.5 months, and the average body mass index was 29.5 kg/m². Hypertension was present in 46% of patients, diabetes in 23%, coronary artery disease in 64%, and obstructive sleep apnea in 46%. The average left atrial diameter was 47.5 mm and the average left ventricular ejection fraction was 29.5%. Baseline 6-minute walk distance averaged 349 m, and the average baseline Minnesota Living with Heart Failure Questionnaire (MLHFQ) score, which ranges from 0-105 with higher scores indicating worse quality of life, was 51.0. Most patients were on guideline-directed heart failure therapy: 90.1% were on an ACE inhibitor or ARB, 78.3% were on a beta-blocker, and 47.8% were on an aldosterone antagonist. Baseline characteristics were similar between the two treatment groups.

Procedure: This was an open-label, randomized, parallel-group, multicenter trial. There was no blinding of patients or treating physicians to treatment assignment, although left ventricular ejection fraction measurements were performed by echocardiographers blinded to randomization group and study endpoints.

Patients were randomized in a 1:1 ratio to catheter ablation or amiodarone.

Patients assigned to catheter ablation underwent pulmonary-vein antrum isolation; the majority (80 of 102 patients, 78.4%) also underwent extension of ablation to achieve isolation of the entire left atrial posterior wall, while the remainder (22 patients, 21.6%) received pulmonary-vein isolation alone, at operator discretion. Additional linear lesions or complex fractionated electrogram ablation were permitted but not mandated. A second ablation procedure was allowed during a 3-month blanking period without being counted as a treatment failure; patients underwent an average of 1.4±0.6 procedures.

Patients assigned to amiodarone received a loading dose of approximately 10 g over the first 2 weeks (400 mg twice daily for 2 weeks), then 400 mg daily for 2 more weeks, followed by a 200 mg/day maintenance dose; digoxin was discontinued or reduced by at least 50% if possible. All patients in both groups received optimal guideline-directed heart failure therapy throughout the study. The authors note that amiodarone itself is also considered a rate-control agent under some guidelines, so this trial should be understood as a comparison of ablation versus a specific antiarrhythmic drug rather than a comparison of ablation versus a dedicated rate-control strategy.

Endpoint: The primary endpoint was long-term procedural success, defined as freedom from AF, atrial flutter, or atrial tachycardia lasting more than 30 seconds while off antiarrhythmic drugs, assessed via device interrogation at 3, 6, 12, and 24 months (arrhythmia recurrence during the first three months were not considered treatment failure). Secondary endpoints included all-cause mortality, AF- and HF-related unplanned hospitalization, change in left ventricular ejection fraction, change in 6-minute walk distance, and change in MLHFQ score, all assessed over a minimum follow-up of 24 months.

The sample size calculation, using a log-rank test, was designed to detect at least a 20% absolute increase in arrhythmia-free success rate (assuming a null hazard rate corresponding to 30-50% success with amiodarone, hazard ratio 0.575) at 24 months, with 80% power and two-sided alpha of 0.05; with 30% oversampling to allow for attrition, 200 patients (100 per group) were required. Notably, this sample size was calculated only for the arrhythmia-recurrence primary endpoint; the trial was not formally powered to detect differences in the secondary hard clinical outcomes of mortality or hospitalization, which should be interpreted with that caveat in mind.

Results: A total of 203 patients were randomized (102 to ablation, 101 to amiodarone); no patients were lost to follow-up. The average follow-up time was 24.0 months for the primary analysis, though patients who underwent a repeat procedure during the blanking period accrued up to 27 months of total follow-up.

At the end of follow-up, 71 patients in the ablation group (70%, 95% CI 60-78%) were free of AF/atrial tachycardia recurrence, compared with 34 patients in the amiodarone group (34%, 95% CI 25-44%; log-rank p<0.001). A sensitivity analysis using a strict 24-month cutoff (rather than allowing extended follow-up after blanking-period re-procedures) showed a similar result: 72% versus 37% (p<0.001). Success rates varied substantially across centers, ranging from 29% to 61% after a single procedure, and were higher with the more extensive posterior-wall isolation approach (79%) than with pulmonary-vein isolation alone (36%; p<0.001). In a multivariable Cox model adjusting for age, sex, diabetes, and hypertension, amiodarone therapy was associated with a significantly higher hazard of arrhythmia recurrence than ablation (HR 2.5, 95% CI 1.5-4.3; p<0.001); diabetes was also independently associated with recurrence.

Unplanned hospitalization over 2 years occurred in 31% of the ablation group versus 57% of the amiodarone group (relative risk 0.55, 95% CI 0.39-0.76; p<0.001; number needed to treat of 3.8). All-cause mortality was lower with ablation: 8 deaths (8%) versus 18 deaths (18%) with amiodarone (log-rank p=0.037). Among the 67 amiodarone-group patients with arrhythmia recurrence, 7 (10.4%) had treatment failure attributed to drug discontinuation for adverse effects (4 thyroid toxicity, 2 pulmonary toxicity, 1 liver dysfunction). Among the 177 surviving patients with end-of-study measurements available, ablation was associated with greater improvement in left ventricular ejection fraction (median +8.3% vs +5.0%, p=0.02), 6-minute walk distance (median +19 m vs +6 m, p=0.02), and MLHFQ score (median improvement of 10 points vs 5 points, p=0.04); improvements were substantially larger among patients who remained arrhythmia-free regardless of treatment assignment. Procedural complications in the ablation group were limited to 2 cases of groin hematoma and 1 pericardial effusion managed conservatively.

Conclusion: In patients with persistent AF, NYHA class II-III heart failure, reduced left ventricular ejection fraction, and an implanted ICD or CRT-D, catheter ablation was superior to amiodarone in achieving freedom from AF at long-term follow-up, and was associated with fewer unplanned hospitalizations and lower all-cause mortality, along with greater improvement in left ventricular ejection fraction, exercise capacity, and quality of life.

Several features of this trial merit caution before extrapolating its mortality and hospitalization findings. First, and most importantly, the trial’s sample size was calculated only around the arrhythmia-recurrence primary endpoint; mortality and hospitalization were secondary, exploratory endpoints, and the trial was not powered for them. The mortality difference (8 vs 18 deaths) rests on a small absolute number of events with a p-value (0.037) close to the conventional threshold; as with any comparison this fragile, only a handful of events would need to shift between groups to overturn statistical significance, and this reduces confidence that the observed hazard reduction reflects a stable, reproducible true effect rather than a chance imbalance in a modestly sized trial. Second, the comparator was amiodarone specifically rather than a dedicated rate-control strategy or an alternative AAD such as dofetilide; this trial therefore establishes that ablation outperforms amiodarone. Third, the trial was open-label for both patients and treating physicians, and while left ventricular ejection assessment was blinded, the softer functional and quality-of-life endpoints (6-minute walk distance, MLHFQ) remain susceptible to the placebo/nocebo and detection-bias concerns inherent to any unblinded comparison, particularly since patients who succeeded with ablation and knew their arrhythmia burden was reduced might reasonably be expected to report and perform better. Fourth, procedural success varied widely by center (29-61%) and was clearly linked to technique (posterior-wall isolation outperforming pulmonary-vein isolation alone), underscoring that these results, like those of PABA-CHF, may not be fully reproducible outside experienced ablation centers using the specific extended-lesion approach used here.

External validity is also worth considering: this population - already carrying an implanted ICD or CRT-D - represents a subset of HF patients with device therapy already in place, of whom only 23.4% of those screened were ultimately randomized, again raising the possibility of a somewhat selected population.

Background: The established paradigm for rhythm control in paroxysmal atrial fibrillation (AF) placed antiarrhythmic drug therapy as first-line treatment, with catheter ablation reserved for patients who failed pharmacological management. By the late 2000s, however, accumulating evidence from single-center studies and early randomized trials, including ThermoCool AF, suggested that pulmonary vein isolation produced superior rhythm control compared with drugs in patients who had already failed at least one antiarrhythmic agent. A parallel hypothesis, grounded in the observation that atrial fibrosis progresses with AF duration and burden, proposed that earlier ablation before structural remodeling had become established might produce better and more durable outcomes than deferred intervention after drug failure. MANTRA-PAF was designed to test this directly: a multicenter randomized trial comparing radiofrequency catheter ablation as first-line treatment with class IC or class III antiarrhythmic drug therapy in patients with symptomatic paroxysmal AF who had never received antiarrhythmic drugs.

The trial was funded by the Danish Heart Foundation with a co-sponsorship contribution from Biosense Webster, the manufacturer of the CARTO mapping system and ThermoCool catheters used in the ablation arm. The steering committee controlled trial design, data collection, analysis, and the decision to publish, with the sponsors explicitly stated to have had no influence on manuscript content - a more independent governance structure than ThermoCool AF, though the Biosense Webster co-sponsorship and multiple investigator disclosures with device industry relationships remain relevant context.

Patients: Eligible patients had symptomatic paroxysmal AF with at least two episodes in the preceding six months, no prior antiarrhythmic drug exposure, and no AF episode exceeding seven days in duration.

Exclusion criteria removed patients older than 70 years, those with contraindications to both class IC and class III agents, prior AF ablation, left atrial diameter exceeding 50 mm, left ventricular ejection fraction below 40%, contraindication to oral anticoagulation, moderate-to-severe mitral valve disease, NYHA class III–IV heart failure, planned structural heart surgery, and secondary AF. The age cap of 70 years and the requirement for absence of any AF episode lasting more than seven days are the two most consequential eligibility features: they define a relatively young population with genuinely paroxysmal - not early persistent - disease, and they exclude the patients most likely to progress despite treatment.

Baseline Characteristics: Between June 2005 and March 2009, 294 patients were enrolled across multiple centers in Denmark, Finland, Germany, and Sweden: 146 randomized to ablation and 148 to antiarrhythmic drug therapy. Mean age was 55±10 years, 70% were male, and mean left atrial diameter was 40 mm in both groups. Left ventricular ejection was above 60% in approximately 80% of patients in each arm. NYHA class I was present in roughly 90%, and CHADS₂ scores were predominantly 0 or 1. Hypertension was present in 29% of the ablation group and 36% of the drug group. Coronary artery disease was uncommon (4% vs. 1%), and structural heart disease was rare. The population is the archetypical low-risk paroxysmal AF patient: young, well-preserved cardiac function, modest comorbidity burden, and small left atria - the cohort in whom both antiarrhythmic drugs and ablation tend to perform at their best.

Procedures: Ablation was performed using radiofrequency energy guided by CARTO electroanatomical mapping, with circumferential encirclement of the left and right pulmonary veins as the procedural goal. Either an irrigated-tip catheter (NaviStar ThermoCool) or a solid 8 mm tip catheter was used, with the goal of eliminating all high-frequency electrical activity exceeding 0.2 mV within the encircled areas. A supplementary linear ablation along the left atrial roof between the two encircled areas was performed routinely; mitral and tricuspid isthmus lines were optional. Notably, the procedural endpoint was elimination of electrical activity within the encircled areas rather than confirmed complete electrical isolation of the pulmonary veins as verified by circular multipolar catheters. Antiarrhythmic medication was permitted during a three-month post-ablation blanking period; supplementary antiarrhythmic drugs were discouraged thereafter, and a second ablation procedure was offered for recurrent AF after the blanking period. Patients in the drug therapy group received flecainide 200 mg/day or propafenone 600 mg/day as first-line agents; class III agents (amiodarone or sotalol) were used if class IC agents were contraindicated. An aggressive rhythm-control strategy including cardioversion and sequential drug trials was recommended for recurrent AF in the drug group, and supplementary ablation was offered if drug therapy failed. The follow-up schedule included 7-day Holter monitoring at 3, 6, 12, 18, and 24 months.

Endpoints: The co-primary endpoints were the cumulative burden of atrial fibrillation, defined as the percentage of total follow-up time spent in AF across all Holter recordings, and the per-visit burden at each individual timepoint. Only AF episodes lasting more than one minute were counted. This burden-based primary endpoint is a continuous measure that captures the total AF experience across follow-up rather than a binary freedom-from-AF endpoint, and it is in principle more sensitive to differences between strategies that affect AF frequency or duration without necessarily eliminating it entirely. Secondary endpoints included freedom from any AF and from symptomatic AF at 24 months, cumulative and per-visit burden of symptomatic AF, time to first recurrence after the blanking period, and SF-36 quality of life at 12 and 24 months.

The sample size calculation was based on an assumed difference in freedom from AF at 24 months of 75% for ablation versus 60% for drug therapy, requiring 150 patients per group for 80% power at a two-sided alpha of 0.05. The primary endpoint (AF burden) was chosen after sample size calculations had already been set, because characteristics of AF burden in paroxysmal AF were unknown when the trial was planned. This sequential disconnect between the power calculation (based on a binary freedom-from-AF endpoint) and the actual primary endpoint (a continuous burden measure) is a structural limitation: the trial was not formally powered for its own primary endpoint.

Results: The primary endpoint of cumulative AF burden did not differ significantly between arms: 90th percentile of arrhythmia burden was 13% in the ablation group and 19% in the drug therapy group (P=0.10). Per-visit burden was also not significantly different at 3, 6, 12, or 18 months. At 24 months, burden was significantly lower in the ablation group (90th percentile 9% vs. 18%, P=0.007), and this was the only timepoint at which the per-visit primary endpoint reached statistical significance. This 24-month signal is important but requires careful contextualisation: it is a single pre-specified timepoint comparison, not the primary result, which was the cumulative measure across all visits. The pattern of non-significant differences across four of five timepoints and a significant difference only at the final visit is consistent with a gradual divergence as drug efficacy wanes over time, a finding that is biologically plausible but not definitive given multiple comparisons without formal multiplicity adjustment.

At 24 months, significantly more patients in the ablation arm were free from any AF (85% vs. 71%, P=0.004) and from symptomatic AF (93% vs. 84%, P=0.01). These secondary endpoint results are the most clinically salient findings in the trial and represent a meaningful absolute difference - 14 percentage points in any-AF freedom and 9 percentage points in symptomatic-AF freedom. They suggest that ablation produces more durable rhythm control than drugs over two years in this population, even though the cumulative burden difference did not reach the primary threshold.

Figure below taken directly from the original publication.

The most clinically consequential finding in the drug arm is the crossover rate: 54 of 148 patients (36%) in the drug therapy group underwent supplementary ablation during the trial, the majority in the first year. This figure is not a failure of the trial design, the protocol permitted crossover as a clinical safety measure, but it is a critical interpretive caveat. The 36% crossover rate means that by the second year of follow-up, a substantial proportion of the drug arm had effectively received the intervention being tested, diluting any between-group difference toward the null and making the intention-to-treat analysis more conservative than the per-protocol comparison. It also reflects the real-world reality that antiarrhythmic drugs fail in a significant minority of patients within the first year. The ablation arm saw a mean of 1.6 procedures per patient, confirming that the majority of the rhythm benefit required more than one ablation.

Quality of life improved significantly from baseline in both groups. The physical component SF-36 summary score improved more in the ablation group over time (P for interaction=0.01), while the mental component showed no between-group difference. This physical component advantage for ablation echoes the ThermoCool AF quality-of-life findings, but in an open-label trial the contribution of expectation and placebo effects cannot be disentangled.

Safety events deserve explicit mention. Three patients in the ablation group developed cardiac tamponade. One patient in the ablation arm died of a procedure-related cerebral stroke. Total serious adverse events were 25 in the ablation arm and 22 in the drug arm (P=0.45), suggesting broadly equivalent overall safety, though the nature of the events differed: the ablation arm carried procedure-related risks (tamponade, stroke, pulmonary vein stenosis) while the drug arm carried drug-related risks (bradycardia requiring pacemaker, hospitalization for heart failure, discomfort attributable to medication).

Conclusions: MANTRA-PAF does not support catheter ablation as routine first-line rhythm control for paroxysmal atrial fibrillation. The primary endpoint (cumulative AF burden over two years) was not significantly different between ablation and drug therapy, and the authors’ own conclusion is that their data support guidelines (at the time) recommending antiarrhythmic drugs as first-line treatment in most patients with paroxysmal AF. This is an honest and appropriately conservative interpretation of the primary result.

The secondary findings at 24 months, lower AF burden, more patients free from any AF and symptomatic AF in the ablation arm, provide a signal that ablation may produce more durable rhythm control than drugs over time, but these are secondary endpoints from a trial that missed its primary endpoint, subject to the usual caveats about multiple comparisons. Whether that trajectory difference translates into clinical outcomes that matter is a question MANTRA-PAF was not designed to answer.

Of note, the procedural technique used - circumferential encirclement targeting electrical silence within the ablation lines rather than confirmed complete pulmonary vein isolation - was already recognized at the time of publication as suboptimal, and the authors explicitly acknowledge that results might have been better with systematic circular multipolar catheter confirmation of electrical disconnection. The mean of 1.6 ablation procedures per patient in the ablation arm further qualifies the efficacy figures: this was not a single-procedure comparison. When read alongside ThermoCool AF which enrolled drug-refractory patients and showed a far larger advantage for ablation, MANTRA-PAF establishes that the benefit of ablation over pharmacological therapy narrows substantially when drug-naïve patients are offered a single antiarrhythmic agent as first-line treatment.

Background: By the mid-2000s, antiarrhythmic drug therapy had established itself as first-line rhythm control option for paroxysmal atrial fibrillation (AF), yet its limitations were well recognized - atrial fibrillation recurrence within six to twelve months approached 50% with most agents, and cumulative adverse effects limited long-term tolerability. Catheter ablation targeting pulmonary vein isolation had emerged as an alternative, supported by single-center and small multicenter studies, but no adequately powered prospective randomized trial had compared ablation with continued antiarrhythmic drug therapy specifically in patients who had already failed at least one drug. ThermoCool AF was designed to fill that gap. It was a multicenter, randomized trial comparing radiofrequency catheter ablation using an irrigated-tip catheter with antiarrhythmic drug therapy in patients with symptomatic paroxysmal AF refractory to initial pharmacological rhythm control.

The trial was funded by Biosense Webster, the manufacturer of the ThermoCool catheter used in all ablation procedures, and was explicitly designed to support FDA market approval of that device. Two Biosense Webster employees are listed as co-authors, the sponsor had overall responsibility for study conduct and data collection, and the sponsor participated in data analysis, interpretation, and manuscript drafting. These relationships are pervasive and represent a substantial conflict of interest that must be held in view throughout the interpretation of results, even though an independent statistical analysis by a Boston University biostatistician confirmed the primary findings.

Patients: Eligible patients were adults with at least three symptomatic AF episodes with one being confirmed by ECG, in the six months before randomization, who had failed to respond to at least one antiarrhythmic drug from class I, class III, or AV nodal blocker category. The requirement for prior drug failure is a critical patient selection feature: it enriches for a population in whom pharmacological rhythm control has already proven inadequate, making the drug arm of the comparison a second attempt at pharmacological therapy rather than a first. This design question — ablation versus re-trial of antiarrhythmic drugs in drug-refractory patients — is clinically meaningful but does not address the prior question of whether ablation should be offered as first-line therapy. Exclusion criteria were extensive and removed the highest-risk and most complex patients: AF duration exceeding 30 days at any episode, left ventricular ejection fraction below 40%, prior AF ablation, amiodarone use within six months, NYHA class III or IV heart failure, recent MI or CABG, prior thromboembolic event within twelve months, severe pulmonary disease, prior valvular surgery, ICD presence, and left atrial diameter of 50 mm or more.

Of 5,545 patients assessed for eligibility, 5,378 were excluded - 4,761 for not meeting inclusion criteria and 617 for refusing participation - leaving 167 randomized. The enrolled population is therefore a narrowly selected group: relatively young, predominantly male, with low cardiovascular comorbidity, preserved LV function, and modestly enlarged left atria.

Baseline Characteristics: A total of 167 patients were randomized 2:1 to catheter ablation (n=106) or antiarrhythmic drug therapy (n=61). Mean age was 56 years, 33.5% were women, and patients had a mean AF history of 5.7 years with an average of 1.3 prior antiarrhythmic drug failures. The two groups were well balanced across all reported baseline variables including age, sex, AF duration, hypertension, diabetes, structural heart disease, NYHA class, left ventricular ejection fraction (mean approximately 62%), left atrial dimension (mean approximately 40 mm), and prior drug failure history. The high LVEF, small left atrial size, low prevalence of structural heart disease (approximately 10–15%), and near-universal NYHA class I status define a population with paroxysmal AF in the context of largely preserved cardiac architecture - the most favorable phenotype for ablation success. Baseline quality of life scores were similar between groups on both the SF-36 and the AF Symptom Frequency and Severity Checklist.

Procedures: Patients randomized to catheter ablation underwent pulmonary vein isolation using the NaviStar ThermoCool irrigated-tip catheter with the CARTO navigation system, with confirmation of entrance block required in all veins. Additional ablation was permitted at investigator discretion and included left atrial linear lesions, ablation at sites of complex fractionated electrograms, and cavo-tricuspid isthmus ablation — applied in 35.9%, 22.3%, and 35.9% of patients respectively. Up to two repeat ablation procedures within 80 days of the initial procedure were allowed; 12.6% of patients underwent a repeat procedure. Anticoagulation with warfarin was required for the first three months post-ablation. Patients randomized to antiarrhythmic drug therapy received a drug not previously administered from the FDA-approved list of dofetilide, flecainide, propafenone, sotalol, or quinidine, at investigator discretion, with amiodarone explicitly excluded. Most patients in the drug arm received flecainide (36%) or propafenone (41%). Following a 3-month blanking period for the ablation arm and a 14-day dose-titration period for the drug arm, both groups entered a comparable 9-month effectiveness evaluation period. Rhythm monitoring was rigorous and symmetric: trans-telephonic monitoring with mandatory weekly transmissions for the first eight weeks then monthly, plus ECGs at all follow-up visits and Holter monitoring at baseline and final visit. Patients in the drug arm were permitted to cross over to ablation after 90 days if treatment failed; 36 of the 47 patients with protocol-defined drug failures did so.

Endpoints: The primary endpoint was time to protocol-defined treatment failure, which encompassed documented symptomatic paroxysmal AF during the effectiveness evaluation period for both groups. Additional criteria for treatment failure in the ablation arm included repeat ablation after day 80, failure to confirm entrance block at the end of the procedure, or changes in specified drug regimen during the evaluation period - meaning that ablation patients could be classified as failures without any AF recurrence, based on procedural or drug-management criteria alone. This asymmetric failure definition arguably biases against the ablation arm by including non-arrhythmic endpoints, though its practical impact on the results appears limited given the magnitude of the between-group difference. The primary statistical framework was prospective Bayesian, with study success defined as a posterior probability of superiority of catheter ablation over drugs of at least 98% at final analysis. Three interim analyses were planned at 150, 175, and 200 patients. There was no formal sample size calculation in the traditional frequentist sense - maximum enrolment was set at 230 patients with a Bayesian adaptive design. The trial stopped at the first interim analysis of 150 patients when the predictive probability of eventual study success exceeded 99%, triggering the pre-specified stopping rule. The subsequent enrolment of the remaining 17 patients - to a total of 167 - occurred before the stopping decision was fully implemented.

Results: At the end of the 9-month effectiveness evaluation period, 66% of patients in the catheter ablation group remained free from protocol-defined treatment failure compared with 16% in the antiarrhythmic drug therapy group (HR 0.30, 95% CI 0.19 - 0.47, P<0.001). The magnitude of this difference - a more than fourfold reduction in treatment failure - is striking. Freedom from symptomatic atrial arrhythmia recurrence was similarly differentiated: 70% with ablation versus 19% with anti-arrhythmic drugs (HR 0.24, 95% CI 0.15 - 0.39). Freedom from any atrial arrhythmia, including asymptomatic events detected by monitoring, was 63% versus 17% (HR 0.29, 95% CI 0.18 - 0.45). These three endpoints tell a consistent story: ablation substantially outperformed continued pharmacological rhythm control across all arrhythmia-related definitions of success.

The quality of life results were equally clear. At three months, ablation patients showed significant improvements in both SF-36 physical and mental component summary scores compared with drug therapy patients, whose scores changed minimally (between-group differences of 6.9 and 6.6 points respectively, both P=0.001). Symptom frequency and severity scores improved substantially in the ablation arm (between-group differences of −11.8 and −9.4 respectively, both P=0.001). These improvements were maintained through the six- and nine-month assessments in the ablation arm. Meaningful comparison in the drug arm was precluded after three months by progressive crossover to ablation.

Major 30-day treatment-related adverse events occurred in 4.9% of ablation patients (one pericardial effusion, one pulmonary oedema, one pneumonia, one vascular complication, one heart failure) and 8.8% of drug therapy patients (two life-threatening arrhythmias, three disabling drug intolerances requiring discontinuation). One patient in the ablation group died 284 days post-procedure from acute myocardial infarction that was judged to be unrelated to the procedure. Notably, none of the most feared ablation complications - thromboembolic events, atrio-esophageal fistula, pulmonary vein stenosis, phrenic nerve paralysis - occurred in this trial, a favorable safety profile the authors partly attribute to the irrigated-tip catheter but which may also reflect the expertise concentrated in the 19 participating centers.

Conclusions: ThermoCool AF establishes the superiority of catheter ablation over continued antiarrhythmic drug therapy in a well-defined, narrowly selected population of patients with symptomatic paroxysmal AF who have failed at least one prior antiarrhythmic drug. The primary finding - 66% versus 16% freedom from treatment failure - is large, consistent across multiple arrhythmia-defined endpoints, and accompanied by clinically meaningful quality-of-life improvements. In this specific population, the trial provides strong evidence supporting catheter ablation as the preferred rhythm-control strategy once pharmacological therapy has failed.

Several important limitations temper how broadly these results can be applied. The population was young, predominantly male, with preserved left ventricular function, near-normal left atrial size, minimal structural heart disease, and low cardiovascular comorbidity - characteristics systematically associated with higher ablation success rates. The exclusion of left atrial diameter ≥50 mm, left ventricular ejection fraction <40%, NYHA class III–IV, and prior persistent atrial fibrillation removes precisely the patients in whom ablation decisions are most contested and outcomes least predictable. The drug arm excluded amiodarone, arguably the most effective antiarrhythmic agent, without clear clinical justification. Sponsor influence at every level of study governance - design, conduct, data collection, analysis, and manuscript drafting - is the most important interpretive caveat; while the independent statistical review provides reassurance, it does not fully neutralize the structural advantages of an industry sponsor designing and executing a trial intended to achieve device market approval. The follow-up of nine months addresses rhythm control but cannot speak to hard outcomes - mortality, stroke, heart failure progression, and quality of life over years rather than months - questions that subsequent larger trials including CABANA would attempt to address.

Finally, this trial - like most AF ablation trials - uses a 90-day blanking period during which arrhythmic events and certain outcomes are excluded from the primary endpoint. Proponents argue that early recurrences do not reliably predict long-term rhythm outcomes and that ablation lesions have not yet fully healed. These are not unreasonable assumptions. However, excluding post-randomization events means that patients and clinicians are not seeing the complete picture. We will revisit this issue in future ablation trial reviews, and our position is consistent: all events after randomization should be reported and transparently communicated.

Background: Catheter ablation for atrial fibrillation (AF) had been shown, in earlier non-randomized and small randomized studies, to improve quality of life, functional capacity and, in retrospective comparisons with antiarrhythmic drugs, survival. Atrioventricular (AV)-node ablation combined with biventricular pacing had separately been used as a rate-control strategy for AF that was refractory to medical therapy, and biventricular pacing had been shown to be superior to right ventricular pacing after AV-node ablation. However, no randomized trial had directly compared pulmonary-vein isolation (a rhythm-control strategy) with AV-node ablation and biventricular pacing (an aggressive rate-control/rate-regularization strategy) in patients with heart failure and reduced ejection fraction. The Pulmonary Vein Antrum Isolation versus AV Node Ablation with Bi-Ventricular Pacing for Treatment of Atrial Fibrillation in Patients with Congestive Heart Failure (PABA-CHF) trial was designed to address this gap.

Patients: Eligible patients had symptomatic, drug-refractory atrial fibrillation, NYHA class II or III heart failure symptoms, and a left ventricular ejection fraction (EF) of 40% or less, while on a stable regimen of beta-blockers and ACE inhibitors (plus spironolactone for NYHA class III patients). Patients had to be able to complete a 6-minute walk test and be 18 years of age or older.

Patients were excluded if they had a reversible cause of AF or heart failure, postoperative AF, previous maze or maze-like surgery, previous left atrial ablation, life expectancy of 2 years or less, high probability of cardiac transplantation within 12 months, contraindication to antiarrhythmic or anticoagulant medications, severe pulmonary disease, an intra-atrial thrombus, tumor or other abnormality precluding catheter placement, or cardiac surgery, myocardial infarction, or percutaneous coronary intervention within the previous 3 months.

Baseline characteristics: Patients were recruited from 13 centers across 6 countries (United States, France, Italy, Australia, Czech Republic, and Germany). Patients were randomized to pulmonary-vein isolation (41 patients) or AV-node ablation with biventricular pacing (40 patients).

The average age of patients was 61 years and 91% were men. Coronary artery disease was present in 70% of patients. Approximately half of patients (52%) had paroxysmal AF, with the remainder having persistent or long-standing persistent AF. The mean duration of AF was approximately 4 years. The average EF was 28%, mean left atrial internal diameter was 4.8 cm, and the average baseline 6-minute walk distance was 275 m. The average baseline Minnesota Living with Heart Failure (MLWHF) questionnaire score, which ranges from 0 to 105 with higher scores indicating worse quality of life, was 89. Baseline characteristics were well balanced between the two groups, with no statistically significant differences.

Procedure: This was an open-label, multicenter, randomized controlled trial; randomization was computer-generated. Both patients and physicians were aware of treatment assignment, although collection of the 6-minute walk distance, echocardiographic measurements, and MLWHF scores was performed by assessors unaware of treatment assignment.

Patients were randomized to pulmonary-vein isolation or AV-node ablation with biventricular pacing.

Patients assigned to pulmonary-vein isolation underwent isolation of all pulmonary-vein antra using a circular mapping catheter and an 8-mm-tip ablation catheter, with additional linear lesions or ablation of complex fractionated electrograms at operator discretion. Warfarin was started after the procedure and continued for at least 3 months (INR goal 2-3), and antiarrhythmic drugs were discontinued at 2 months but could be restarted thereafter at physician discretion. A second pulmonary-vein isolation or ablation for atrial flutter was permitted at 3 months for recurrent arrhythmia.

Patients assigned to AV-node ablation underwent ablation of the bundle of His to achieve complete heart block, followed by implantation of a biventricular ICD. Crossover between assigned strategies was discouraged during the study period, although patients in the pulmonary-vein isolation group could receive an ICD for standard primary or secondary prevention indications, and crossover to the other strategy was permitted after the 6-month study period ended.

Endpoint: The primary endpoint was a prespecified composite of left ventricular ejection fraction, distance on the 6-minute walk test, and MLWHF score at 6 months. Because this was a hierarchically tested composite, the trial’s statistical plan (borrowed from the MIRACLE and MIRACLE-ICD trials) declared the primary endpoint met if all three components had a p-value <0.05, if any two had p<0.025, or if any single component had p<0.017. Secondary endpoints were freedom from AF and left atrial internal diameter at 6 months, each considered significant at the conventional p<0.05 threshold.

The sample size calculation estimated that 40 patients per group would provide 80% power (two-sided alpha of 0.017) to detect a between-group difference of 35 m in 6-minute walk distance, 6% in ejection fraction, and 15 points in MLWHF score. This is a modest sample size for a trial with three co-primary components, and the graduated significance thresholds (allowing significance to be declared even if only one of three components reaches a stringent p-value) make the composite considerably easier to satisfy than a single hard endpoint tested at conventional alpha.

Results: All 81 randomized patients (41 to pulmonary-vein isolation, 40 to AV-node ablation with biventricular pacing) completed follow-up; none were lost at 6 months. In the pulmonary-vein isolation group, 8 patients (19.5%) underwent a repeat procedure for recurrent AF, atrial flutter, or atrial tachycardia.

At 6 months, all three components of the composite primary endpoint favored pulmonary-vein isolation: ejection fraction was 35.0±9.0% versus 28.0±6.0% (p<0.001), 6-minute walk distance was 340.0±49.0 m versus 297.0±36.0 m (p<0.001), and MLWHF score was 60.0±8.0 versus 82.0±14.0 (p<0.001). Left atrial internal diameter decreased with pulmonary-vein isolation (4.9±0.5 cm to 4.5±0.4 cm) and increased slightly with AV-node ablation (4.7±0.6 cm to 4.9±0.6 cm; p<0.001 for between-group comparison).

Freedom from AF (with or without antiarrhythmic drugs) occurred in 88.0% of the pulmonary-vein isolation group at 6 months, and in 71.0% without any antiarrhythmic drug use; by definition, 0% of the AV-node ablation group were free from AF, since the AV node itself had been ablated and the atria remained in fibrillation. Progression of AF (worsening from paroxysmal toward persistent) occurred in 0% of the pulmonary-vein isolation group versus 30.0% of the AV-node ablation group (p<0.001). Complication rates were similar between groups (groin bleeding, pericardial effusion, and mild pulmonary-vein stenosis in the ablation group; lead dislodgment, elevated pacing threshold, pocket hematoma, and pneumothorax in the AV-node ablation group); no deaths occurred during either procedure or follow-up. The average follow-up time was 6.0 months for all patients.

Conclusion: In patients with drug-refractory AF, NYHA class II-III heart failure, and an EF of 40% or less, pulmonary-vein isolation produced greater improvements in ejection fraction, functional capacity, and quality of life at 6 months than AV-node ablation with biventricular pacing, and largely eliminated AF without antiarrhythmic drugs in the majority of patients.

Several features of this trial limit how far its results can be extended. It was small (81 patients), open-label, and unblinded for every component of its primary endpoint, and two of the three components (MLWHF score and, arguably, the clinician-influenced decision to restart antiarrhythmic drugs) are soft, subjective measures that are particularly susceptible to placebo/nocebo effects and to detection bias when both patients and treating physicians know the assignment. Only the EF and, to a lesser extent, the 6-minute walk distance are relatively objective; even so, an unblinded rhythm-control strategy that restores sinus rhythm is expected on physiological grounds to raise EF, so this comparison does not equally test “ablation versus AV-node ablation” so much as it tests “sinus rhythm versus persistent AF with a regularized ventricular rate.” The trial was also not designed or powered for hard clinical outcomes such as mortality or heart failure hospitalization, and no such data were reported. Additionally, the composite primary endpoint used a graduated significance schema that made it easier to declare a positive result than if each component required conventional two-sided alpha of 0.05 individually.

External validity is another important limitation. The trial was selective, which is understandable for a procedural trial. All procedures were performed at experienced ablation centers by high-volume operators, a caveat the authors themselves acknowledge.

PABA-CHF was an important proof-of-concept trial showing that pulmonary-vein isolation produces superior morphologic, functional and quality-of-life outcomes compared with AV-node ablation and biventricular pacing in a highly selected group of heart failure patients with drug-refractory AF. However, given its open-label design, reliance on soft and semi-subjective endpoints, small sample size, and lack of hard clinical outcomes, its findings are best viewed as hypothesis-generating for the subsequent generation of trials.

Since this trial, larger and better-powered randomized trials have directly examined hard outcomes with AF ablation in heart failure with reduced EF (CASTLE-AF and CASTLE-HTx) which we will review in the upcoming weeks.

If you find Cardiology Trials useful, the most valuable thing you can do is share it with a colleague who thinks carefully about trial evidence. Referrals unlock access to paid content — and help keep the publication going.

How to participate

1. Share Cardiology Trial’s Substack. When you use the referral link below, or the “Share” button on any post, you'll get credit for any new subscribers. Simply send the link in a text, email, or share it on social media with friends.

Refer a friend

2. Earn benefits. When more friends use your referral link to subscribe (free or paid), you’ll receive special benefits.

• Get a 1 month comp for 3 referrals

• Get a 3 month comp for 5 referrals

• Get a 6 month comp for 10 referrals

Visit the leaderboard

To learn more, check out Substack’s FAQ.

Thank you for helping get the word out about Cardiology Trial’s Substack!

Background: Patients who present to the emergency department with recent-onset, symptomatic atrial fibrillation have long been treated with prompt restoration of sinus rhythm, either pharmacologically or electrically. The reflex is intuitive: the patient is symptomatic, the arrhythmia is visible on the monitor, and cardioversion makes it disappear. But the rationale for immediate cardioversion had never been rigorously established, and it sits uneasily against a well-documented fact that recent-onset atrial fibrillation frequently terminates on its own. Ask any cardiologist or cardiology trainee, and they will tell you that it is not uncommon for patients scheduled for cardioversion to revert to sinus rhythm before the procedure can be performed.

If a large fraction of patients will convert spontaneously within a day or two, then immediate cardioversion subjects many of them to a procedure, and its attendant sedation and small but real risks, that they never needed. The competing approach is a wait-and-see strategy: control the rate, treat the symptoms, send the patient home, and cardiovert only if the arrhythmia has not resolved within 48 hours. RACE 7 ACWAS was designed to test, as a noninferiority comparison, whether this delayed strategy achieves as much sinus rhythm at four weeks as immediate cardioversion. It is worth noting at the outset that the trial’s endpoint is a rhythm-status snapshot, not a clinical-outcome comparison, which bounds what the trial can establish.

Patients: Eligible patients were adults presenting to the emergency department with hemodynamically stable, symptomatic, recent-onset atrial fibrillation of less than 36 hours’ duration, either first-detected or recurrent, without signs of myocardial ischemia and without a history of persistent atrial fibrillation (defined here as an episode lasting more than 48 hours). All had to be candidates for either strategy. The exclusion list is clinically sensible: Hemodynamic instability, pre-excitation, sick sinus syndrome, unexplained syncope, acute heart failure - but it also defines the boundary of generalizability: this is a stable, uncomplicated, early-presenting population, and the results do not extend to the unstable patient, the patient with atrial fibrillation of unknown or longer duration, or the patient in whom rate control is untenable. The screening log from two centers is telling: of 3,706 patients screened, 2,581 (70%) were not eligible, most commonly for atrial fibrillation duration over 36 hours or a history of persistent atrial fibrillation.

Baseline Characteristics: The trial randomized 437 patients across 15 Dutch hospitals (3 academic, 8 nonacademic teaching, 4 nonteaching), with 218 assigned to delayed cardioversion (wait-and-see) and 219 to early cardioversion. The mean age was 65 years and 40% were women. Atrial fibrillation was a first detected episode in 44%. The median heart rate during atrial fibrillation was approximately 124 bpm, and palpitations were the dominant symptom (87%), followed by exercise-induced fatigue, dyspnea, and chest pain. A total of 64% had a CHA₂DS₂-VASc score of 2 or higher, hypertension was present in 57%, diabetes in 11%, prior myocardial infarction in 8%, and prior ischemic stroke or TIA in 6%. At enrollment, 40% were already on an oral anticoagulant and about a fifth were on an antiarrhythmic drug.

Procedures: The trial was investigator-initiated, multicenter, randomized, and open-label, with central assessment of the primary-endpoint ECGs by the first two authors. Randomization was 1:1 via a centralized web-based system; patients and physicians were aware of assignment.

In the wait-and-see arm, patients received rate-control medication (predominantly metoprolol; beta-blockers in the large majority, with calcium-channel blockers or digoxin used less often) titrated to symptom relief and a heart rate of 110 bpm or less, then were discharged and reassessed at an outpatient visit timed as close as possible to 48 hours after symptom onset. If still in atrial fibrillation, they were referred for delayed cardioversion. In the early-cardioversion arm, patients underwent immediate cardioversion, pharmacologic (preferably flecainide) by preference and electrical when pharmacologic conversion was contraindicated or had failed. Anticoagulation was managed by guideline-based CHA₂DS₂-VASc criteria, initiated before or immediately after cardioversion in high-risk patients not already anticoagulated; notably, transesophageal echocardiography was not performed in any patient, consistent with the <48-hour window in which left atrial thrombus is presumed unlikely. Follow-up ECG telemetry using MyDiagnostick, Applied Biomedical Systems (three times daily and with symptoms) was used to detect recurrences, but devices were unavailable for 102 patients.

The trial was supported by the Netherlands Organization for Health Research and Development and Maastricht University Medical Center. Boehringer Ingelheim provided some of the remote monitoring devices but had no role in design or execution and did not review the protocol or manuscript.

Endpoints: The primary endpoint was the presence of sinus rhythm on the ECG recorded at the 4-week visit.

The noninferiority margin was prespecified as a lower 95% confidence bound for the between-group difference no worse than −10 percentage points, equivalent to one-sided testing at alpha 0.025.

The objective was to show the noninferiority of a wait-and-see (delayed-cardioversion) approach compared with early cardioversion for the presence of sinus rhythm at 4 weeks.

Results: Sinus rhythm at four weeks was present in 193 of 212 patients (91%) in the delayed-cardioversion group and 202 of 215 (94%) in the early-cardioversion group, a between-group difference of −2.9 percentage points (95% CI −8.2 to 2.2; P= 0.005 for noninferiority). The lower confidence bound of −8.2 sits inside the −10 margin, so noninferiority is met. The post-hoc sensitivity analyses including all randomized patients (Table S3) produced similar estimates (−3.2 and −3.7 points), so the result is not an artifact of the 10 excluded patients.

The mechanism behind the result is its most illuminating feature. In the wait-and-see arm, 69% of patients converted spontaneously within 48 hours on rate-control medication alone, and only 28% required delayed cardioversion. In the early-cardioversion arm, conversion was spontaneous in 16% before the procedure could even begin and followed cardioversion in 78%. In other words, a large majority of the delayed-strategy patients never needed a cardioversion at all, and a meaningful minority of the immediate-strategy patients were cardioverted when they would have converted on their own had anyone waited. This is the clinical heart of the trial: immediate cardioversion treats a problem that frequently resolves itself.

Figure below taken from the NEJM publication.

Recurrence of atrial fibrillation was common and essentially identical between strategies. Approximately 7% in each arm visited the emergency department because of atrial fibrillation recurrence. Among the 335 patients with telemetric monitoring, recurrence within four weeks occurred in 30% of the delayed group and 29% of the early group, with a hazard ratio of 0.97 (95% CI 0.65 to 1.43) and Kaplan–Meier curves that overlap throughout. This is an important and underemphasized point: whatever a clinician does acutely, roughly a third of these patients will have a recurrence within a month, and the acute strategy does not change that. Restoring sinus rhythm quickly does not buy durable rhythm stability, it merely changes the route to the same four-week destination.

Cardiovascular complications were infrequent and not significantly different (10 in the delayed group, 8 in the early group), with no deaths. The trial was explicitly not powered for safety, so these numbers are reassuring but not conclusive, and the events are worth reading individually rather than in aggregate.

Conclusions: In patients presenting to the emergency department with stable, recent-onset, symptomatic atrial fibrillation, a wait-and-see approach with rate control and delayed cardioversion only if needed was noninferior to immediate cardioversion for the presence of sinus rhythm at four weeks. Several aspects deserve careful consideration.

First, the noninferiority margin of −10 percentage points is wide. Nevertheless, the authors provide a reasonable justification, stating: “A noninferiority margin of 10 percentage points was considered acceptable, given the natural variation in the presence of sinus rhythm, the generally low effect of the absence of sinus rhythm on prognosis of the patient, and the availability of good treatment options should treatment be necessary”. In our view, this is a reasonable justification given the nature of the endpoint. However, such a wide noninferiority margin would not have been acceptable for hard clinical outcomes such as mortality, hospitalization, myocardial infarction or stroke.

Second, the applicability of the findings is limited by the highly selected study population. Because 70% of screened patients were not eligible, the results apply primarily to stable patients with recent-onset, uncomplicated atrial fibrillation.

Third, the endpoint is a rhythm snapshot, not a clinical outcome. The trial was not powered for, and does not establish, noninferiority for stroke, heart failure or death. The near-identical (around 30%) four-week recurrence rates in both arms reinforce that the acute decision does not determine the medium-term arrhythmia course.

Despite these limitations, the trial addresses a highly relevant question faced daily by emergency physicians and cardiologists, and the authors should be commended for conducting it. The trial was published in 2019 and the findings are even more important nowadays as increasing numbers of patients present with newly detected atrial fibrillation, often prompted by wearable and digital device alerts, sometimes in the absence of symptoms.

Several numbers are worth remembering. In the wait-and-see group, 69% of patients converted spontaneously within 48 hours with rate-control therapy alone, and only 28% ultimately required cardioversion. Even in the early-cardioversion group, 16% converted spontaneously before the procedure could be performed. Moreover, 20% of otherwise eligible patients were never randomized because they had already returned to sinus rhythm. Together, these findings suggest that many patients can avoid an unnecessary procedure, with important implications for healthcare utilization, resource allocation and cost.

Background: For patients with infrequent, well-tolerated episodes of paroxysmal atrial fibrillation, the standard management options in the early 2000s carried a mismatch between the burden of treatment and the burden of disease. Daily antiarrhythmic prophylaxis imposed continuous drug exposure and its attendant toxicity on patients whose atrial fibrillation was episodic and self-limited. Catheter ablation was emerging but reserved for patients with frequent episodes refractory to pharmacological prophylaxis. Emergency room visits and hospitalizations for cardioversion, meanwhile, represented a substantial personal and healthcare cost for a condition whose individual episodes were often brief and hemodynamically tolerated. The class IC agents flecainide and propafenone had been established through multiple in-hospital placebo-controlled trials as effective oral cardioversion agents, with conversion rates of 58–95% within six to eight hours of administration. The pill-in-the-pocket concept (patient-initiated self-administration of a single oral loading dose at episode onset) offered a potential middle path between continuous prophylaxis and repeated emergency presentations. This study set out to evaluate whether that approach was feasible and safe when deployed outside the hospital in a carefully selected population. It was funded by an Italian cardiology society with no pharmaceutical industry involvement declared.

Patients: Eligibility was narrow and deliberately so. Patients had to be aged 18–75, present to the emergency room with ECG-documented atrial fibrillation of less than 48 hours’ duration, have a mean heart rate above 70 bpm and systolic blood pressure of at least 100 mmHg, and report a history of palpitations with abrupt onset that were hemodynamically well tolerated; no dyspnea, presyncope, or syncope. They were required to have had at least one but fewer than 12 episodes in the preceding year and to have no cardiac symptoms outside of arrhythmic episodes. The exclusion list was extensive and clinically meaningful: ventricular pre-excitation, bundle branch block, ischemic heart disease, dilated or hypertrophic cardiomyopathy, any history of heart failure, severe valvular disease, left ventricular ejection fraction below 50%, long QT syndrome, Brugada syndrome, bradycardia-tachycardia syndrome, prior AV block, prior thromboembolism, renal or hepatic insufficiency, prior hypokalemia, and any current antiarrhythmic prophylaxis. These criteria define the intended population with precision: structurally normal or mildly diseased hearts, infrequent paroxysmal atrial fibrillation, good hemodynamic tolerance, and no contraindications to class IC agents. At the two sites where all-comers with recent-onset atrial fibrillation were prospectively tracked, only 12% of patients presenting to the emergency room met criteria for out-of-hospital treatment, a figure that anchors the applicability of this strategy in realistic perspective.

Baseline Characteristics: Of 268 patients treated in hospital with flecainide or propafenone, 58 (22%) were excluded from the out-of-hospital phase: 41 (15%) due to drug failure to convert within six hours, 14 (5%) due to in-hospital adverse effects including transient atrial flutter in seven and hypotension in four, and three due to echocardiographic exclusion criteria identified only after admission. The remaining 210 patients were enrolled for out-of-hospital treatment. Their mean age was 59 years, 58% were men, and 44% had mild heart disease. Mean left ventricular ejection fraction was 59% and mean left atrial diameter was 39 mm, confirming the preserved cardiac phenotype targeted by the protocol. History of atrial fibrillation was a mean of four years, with a median of 3.3 symptomatic episodes in the prior year and 2.7 emergency room contacts per patient. The two-stage selection process - in-hospital drug testing followed by out-of-hospital enrolment - means that the 210 patients who received the pill-in-the-pocket prescription were pre-screened responders who had already demonstrated both drug efficacy and absence of significant adverse effects. This is a critical design feature: it is simultaneously a safety strength and a limitation on generalizability.

Procedures: The trial used a two-stage enrollment process. In the first stage, patients were enrolled for in-hospital treatment if they met all inclusion criteria and had no documented exclusion criteria. In the second stage, patients who had been successfully treated during hospitalization and continued to meet eligibility criteria on subsequent evaluation were enrolled for out-of-hospital treatment.

Patients were instructed to take their assigned drug five minutes after any subsequent onset of palpitations, in weight-adjusted loading doses (flecainide 300 or 200 mg or propafenone 600 or 450 mg for patients weighing ≥70 kg or <70 kg, respectively). After ingestion, a resting supine or sitting position was recommended until palpitations ceased or four hours had elapsed. Patients were given a diary to record episode timing, drug ingestion, symptom resolution, and adverse effects. They were advised to contact the emergency room if palpitations persisted beyond six to eight hours, if new symptoms emerged, or if they felt a marked acceleration of heart rate after drug ingestion - the latter being the clinical signal for atrial flutter with rapid AV conduction. Only one dose per 24-hour period was permitted. Outpatient clinic review occurred every four months. The choice of drug at each center reflected local familiarity rather than randomization, a pragmatic decision that precludes any head-to-head comparison between the two agents. No placebo group was included, a choice the authors justify on the grounds that in-hospital superiority of both drugs over placebo had already been established and that the primary question was feasibility and safety rather than comparative efficacy - a reasonable position, though it means the study cannot quantify spontaneous conversion rates in this out-of-hospital setting.

Endpoints: The primary endpoints were the rate of treatment success defined as resolution of palpitations within six hours of drug ingestion, and the rate of adverse events during out-of-hospital treatment. Secondary endpoints were rates of emergency room visits and hospitalizations during follow-up compared with the year before enrollment. No formal sample size calculation was reported, which is consistent with the study’s framing as a feasibility and safety evaluation rather than a hypothesis-testing trial powered for a pre-specified effect size.

Results: Over a mean follow-up of 15 months, 165 of the 206 patients with complete follow-up (79%) had at least one atrial fibrillation recurrence, generating 618 episodes in total. Of those, 569 (92%) were treated with the pill-in-the-pocket drug, a median of 10 minutes after symptom onset. Treatment was successful (palpitations resolved within six hours) in 534 of 569 treated episodes (94%; corrected efficacy 93%, 95% CI 90–95%). Mean time to symptom resolution was 113 ± 84 minutes. Flecainide and propafenone performed identically at 94% each. In 139 of the 165 patients with recurrences (84%), the drug was effective across all episodes experienced during follow-up.

Adverse events during out-of-hospital treatment occurred in 12 patients (7%). The clinically significant event was a single case of atrial flutter with 1:1 atrioventricular conduction and a ventricular rate of 210 bpm in a patient who had previously tolerated in-hospital flecainide without incident (this is a reminder that successful in-hospital testing does not immunize against this complication on subsequent exposures). The remaining 11 patients reported non-cardiac effects including nausea, asthenia, and vertigo. No ventricular arrhythmias, syncope, or deaths were reported during out-of-hospital treatment. Twenty-seven patients (13%) dropped out during follow-up: 13 due to drug inefficacy, 7 due to frequent recurrences necessitating prophylactic therapy, 2 due to adverse effects, and 5 due to psychological difficulties or non-compliance.

Emergency room contact during follow-up totaled 31 episodes (5% of all 618 episodes). Monthly emergency room visits fell from 45.6 in the year before enrollment to 4.9 during follow-up (P<0.001), and monthly hospitalizations fell from 15.0 to 1.6 (P<0.001). These reductions are numerically striking but must be interpreted cautiously. The before-after design conflates the drug effect with secular trends in patient behavior, regression to the mean, and the psychological reassurance of having a home treatment available. That 11 of the 31 emergency room contacts during follow-up were attributable to anxiety rather than hemodynamic compromise adds a further note of complexity.

Conclusions: The pill-in-the-pocket study establishes proof of concept for patient-initiated oral cardioversion in a carefully selected low-risk population and does so with a safety profile that, within the study’s limitations, is reassuring. The 94% episode-level efficacy and the substantial reduction in emergency room visits are clinically compelling and have appropriately anchored this approach in international atrial fibrillation guidelines as an option for suitable patients.

The critical caveat is that the population enrolled here is not the atrial fibrillation population in general, it is a pre-screened, responder-enriched cohort representing approximately 12% of all patients presenting to the emergency room with recent-onset atrial fibrillation. The two-stage in-hospital test-dose design is both the study’s most important safety feature and its most important limitation: it systematically excludes the 22% of patients in whom the drug fails or causes adverse effects at first exposure, then reports the out-of-hospital experience only in proven responders. This is not a criticism of the design, but it means that the 94% efficacy and 7% adverse event rate observed out of hospital are conditional on prior in-hospital success and cannot be extrapolated to unselected patients or to first-time use without preceding in-hospital testing.

The absence of a control arm and the before-after design for the secondary endpoints further limit what can be formally concluded about the magnitude of healthcare utilization reduction attributable to the drug, as opposed to patient behavior change, natural history or regression to the mean.

The before-after comparison is vulnerable to multiple confounders, and the authors acknowledge they cannot exclude that some treated episodes were not atrial fibrillation at all, since rhythm confirmation during out-of-hospital episodes relied on patient-reported palpitation cessation rather than ECG documentation. The single case of atrial flutter with 1:1 AV conduction out of hospital - a potentially life-threatening arrhythmia - occurred despite successful in-hospital drug testing and serves as a reminder that class IC agents carry a small but irreducible proarrhythmic risk that cannot be fully screened for in advance. The practical implication is that co-administration of an AV nodal blocking agent should be considered in patients without contraindications, particularly those with higher resting heart rates, even if this was not systematically required in the study protocol. Taken together, this is an honest and appropriately scoped feasibility study that defines the right patient for pill-in-the-pocket therapy rather than making broad claims about universal applicability.

Background: Atrial fibrillation and heart failure frequently coexist. Observational data from SOLVD, Framingham, and multiple registries consistently identified atrial fibrillation as an independent predictor of mortality in patients with left ventricular dysfunction, generating a plausible but untested hypothesis that suppressing atrial fibrillation should improve survival. Six trials, including AFFIRM and RACE, had already failed to demonstrate a mortality benefit from rhythm control over rate control in patients with atrial fibrillation, but fewer than 10% of participants in those trials had significant left ventricular dysfunction. The question of whether the heart failure population was different - whether rhythm control might confer a survival advantage specifically in patients with impaired systolic function - remained genuinely open. AF-CHF was designed to answer it directly. The trial was funded by the Canadian Institutes of Health Research with pharmaceutical companies providing study medications at no charge to uninsured patients; the principal investigator and coordinating center were academic, and the disclosures, while extensive, reflect consulting rather than ownership relationships. Compared to many industry-sponsored trials in this space, the governance structure is relatively arm’s-length.

Patients: Eligible patients required a left ventricular ejection fraction ≤35% measured within six months of enrollment, a history of symptomatic heart failure (NYHA class II–IV within the prior six months, hospitalization for heart failure within six months, or LVEF ≤25% regardless of symptoms), and documented atrial fibrillation defined as at least one episode lasting six hours or requiring cardioversion within the prior six months. These inclusive atrial fibrillation criteria are worth noting: they allowed enrollment of patients with paroxysmal, persistent, or recently terminated atrial fibrillation, provided ECG documentation existed. Exclusion criteria included persistent atrial fibrillation for more than 12 months. Other exclusions were decompensated heart failure within 48 hours, bradycardia below 50 bpm, long QT syndrome, dialysis dependence, and anticipated transplant within six months. Enrollment was open and unblinded, which introduces the possibility of investigator-level selection bias, though the stratified randomization scheme mitigates this at the group level. The 123-center international design spanning Canada, the United States, Brazil, Argentina, and Europe improves generalizability but introduces heterogeneity in practice patterns and medication access.

Baseline Characteristics: A total of 1,376 patients were enrolled: 682 to rhythm control and 694 to rate control. The mean age was 67 years and 82% were men. Mean left ventricular ejection fraction was 27±6%. Coronary artery disease was the predominant etiology in 48%, nonischemic cardiomyopathy in approximately 37%. Persistent atrial fibrillation was present in approximately 69%, and more than half of patients had previously been hospitalized for either atrial fibrillation or heart failure, underscoring the high-risk, symptomatic nature of the cohort. Background therapy was modern and appropriate for the era: beta-blockers in 79%, ACE inhibitors or ARBs in 97%, aldosterone antagonists in approximately 44%, and oral anticoagulants in 88%. Implantable cardiac defibrillator use was low at 7% at baseline, rising to approximately 16% during follow-up. The two groups were broadly well matched, with a small but potentially meaningful imbalance in male sex (78% vs. 85% in the rhythm- and rate-control groups respectively) and a slightly higher proportion of patients in atrial fibrillation at the time of ECG in the rate-control group (61% vs. 54%), which could have influenced how aggressively rhythm control was pursued in the early weeks.

Procedures: The rhythm-control strategy called for electrical cardioversion within six weeks of randomization in patients not converting spontaneously on antiarrhythmic drug therapy, with repeat cardioversion recommended for recurrences. Amiodarone was the mandated first-line agent for rhythm maintenance; sotalol or dofetilide were permitted alternatives. At 12 months, 82% of the rhythm-control group were taking amiodarone, declining modestly to 73% by 36 months. Rate control targeted a resting ventricular rate below 80 bpm on ECG and below 110 bpm on a 6-minute walk test, achieved with beta-blockers and digoxin; AV nodal ablation with pacing was available for refractory cases. These rate targets were achieved in 82–88% of the rate-control group over the first three years. The trial was open-label - patients and physicians knew their assignment. Clinical events were adjudicated by an independent external events committee blinded to treatment allocation, providing some protection against ascertainment bias in the primary endpoint. A data safety monitoring board reviewed outcomes twice yearly; six pre-specified interim analyses were performed, each at a stringent alpha of 0.00014, yielding a final significance threshold of 0.04998.

Endpoints: The primary endpoint was death from cardiovascular causes. Secondary endpoints included all-cause mortality, stroke, worsening heart failure (defined as requiring hospitalization, intravenous diuresis, or a change in treatment strategy), the composite of cardiovascular death, stroke, or worsening heart failure, and quality of life and cost. The secondary endpoint hierarchy was pre-specified and the composite was a logical extension of the primary. One structural concern is that worsening heart failure, as a secondary endpoint, was defined broadly enough to include changes in treatment strategy, a partially subjective determination that may have been influenced by knowledge of the assigned strategy in this open-label trial.

The original sample size calculation assumed a two-year cardiovascular mortality rate of 19% in the rate-control group, with 80% power to detect a 25% relative reduction in the rhythm-control group, a two-year accrual period, four years of total follow-up, and 2% annual loss to follow-up - yielding an initial target of 1,450 patients. When recruitment extended beyond the original timeline, the calculation was revised in 2005: the longer-than-anticipated follow-up meant that 1,374 patients would suffice to maintain the same statistical power. The final enrolled population of 1,376 met the revised target. The assumed 25% relative reduction threshold is worth flagging: it is a substantial effect size, and a trial powered to detect it will be underpowered for more modest but still clinically meaningful reductions.

Results: The primary endpoint was resoundingly neutral. Cardiovascular death occurred in 182 patients (27%) in the rhythm-control group and 175 (25%) in the rate-control group (HR: 1.06, 95% CI 0.86–1.30, P=0.59). The Kaplan-Meier curves overlapped from the outset and showed no tendency to diverge at any timepoint across a mean follow-up of 37 months. After adjustment for pre-specified baseline covariates the hazard ratio was essentially unchanged at 1.05 (95% CI 0.85–1.29). All-cause mortality was 32% versus 33% (HR 0.97, 95% CI 0.80–1.17, P=0.73). Stroke occurred in 3% versus 4% (HR 0.74, 95% CI 0.40–1.35, P=0.32). Worsening heart failure was 28% versus 31% (HR 0.87, 95% CI 0.72–1.06, P=0.17). The composite of cardiovascular death, stroke, or worsening heart failure was 43% versus 46% (HR 0.90, 95% CI 0.77–1.06, P=0.20). None of 10 pre-specified subgroups including age, NYHA class, ejection fraction stratum, atrial fibrillation type, and beta-blocker use showed heterogeneity of effect. The consistency is striking and leaves little room for a meaningful subgroup that might benefit from pharmacological rhythm control.

Figure below taken directly from the primary publication in NEJM.

The rhythm-control strategy did achieve its intermediate goal: at follow-up visits, 73–83% of rhythm-control patients were in sinus rhythm on ECG, compared with 30–41% in the rate-control group. Yet 58% of rhythm-control patients had at least one recurrence of atrial fibrillation during follow-up, and 21% crossed over to rate control - most commonly because sinus rhythm could not be maintained. This is the core dilemma the trial illuminates: even with aggressive amiodarone-based rhythm control, imperfect suppression of atrial fibrillation in a high-risk population with structural heart disease may be insufficient to alter the biological trajectory of the disease. The authors explicitly raise the possibility that atrial fibrillation in heart failure may be a marker of disease severity rather than an independent driver of prognosis - a confounded association rather than a causal one - and the trial result is consistent with this interpretation.

The rate-control group had lower rates of hospitalization overall in the first year (39% vs. 46%, P=0.001), driven by fewer admissions for atrial fibrillation management (9% vs. 14%, P<0.001) and fewer bradyarrhythmia admissions (3% vs. 6%, P=0.02). The rhythm-control group required substantially more electrical cardioversions (59% vs. 9%). Torsade de pointes occurred in one patient in the rhythm-control group and none in the rate-control group. Rates of sustained ventricular tachyarrhythmia and major non-non-cranial hemorrhage were similar between groups.

Conclusions: AF-CHF closes the question that AFFIRM left open: even in the specific population of patients with reduced ejection fraction and symptomatic heart failure, a pharmacological strategy of rhythm control provides no mortality benefit over rate control. The consistency of the null result across all subgroups and all secondary endpoints makes this a genuinely informative trial rather than a merely underpowered one.

The most important caveat is what this trial actually tested: rhythm control meant amiodarone in 82% of cases. Amiodarone achieves sinus rhythm imperfectly - 58% of rhythm-control patients had at least one atrial fibrillation recurrence - and carries its own toxicity burden, including thyroid, pulmonary, and hepatic effects that accumulate with prolonged use and that may themselves contribute adversely to outcomes in a vulnerable population. The question of whether a more effective and less toxic rhythm-control intervention - specifically catheter ablation - might produce different results is one AF-CHF was neither designed nor powered to answer. The subsequent CASTLE-AF and AATAC trials would later suggest that ablation-based rhythm control may improve outcomes in heart failure with reduced ejection fraction, making AF-CHF less a refutation of the rhythm-control concept than a refutation of pharmacological rhythm control as practiced in the mid-2000s.

The overall mortality rate of approximately 10% per year in both arms was higher than the rate-control mortality assumed in the original power calculation (19% at two years), a discrepancy that prompted the sample size revision in 2005 and reflects the genuinely high-risk nature of this population. The low ICD implantation rate at baseline - 7% in a population with mean left ventricular ejection fraction of 27% - is striking by contemporary standards, and the authors acknowledge that broader ICD use might have reduced mortality differentially in the rhythm-control arm, where amiodarone’s anti-arrhythmic coverage was imperfect and 36% of all deaths were presumed arrhythmic. That approximately one in three deaths was arrhythmic despite antiarrhythmic drug therapy in the rhythm-control arm reinforces both the limitations of pharmacological approaches to sudden death prevention and the underutilization of ICD therapy in this trial population. Taken together, AF-CHF is a methodologically sound, adequately powered trial with a clean primary result that appropriately established rate control as the default strategy for pharmacological management of atrial fibrillation in heart failure with reduced ejection fraction.

Background: By 2000, dofetilide had already been studied in the high-risk DIAMOND-CHF population, where it demonstrated mortality neutrality and pharmacological cardioversion efficacy as a secondary substudy finding. SAFIRE-D was designed as the dedicated efficacy trial: a dose-ranging study purpose-built to quantify dofetilide’s ability to convert chronic atrial fibrillation or flutter to sinus rhythm and to maintain it over a full year, in a population not selected for heart failure or severe left ventricular dysfunction. This is a meaningful shift in study population and intent relative to DIAMOND-CHF - SAFIRE-D asks an efficacy question in a broader atrial fibrillation population, whereas DIAMOND-CHF asked a safety question in a high-risk heart failure population. The trial was funded entirely by Pfizer, the drug’s manufacturer, and a Pfizer employee is listed as a co-author. As with the other dofetilide trials in this manufacturer’s program, this sponsorship relationship is worth holding in mind when interpreting the framing of results, though the use of an independent statistical methodology (Cochran-Mantel-Haenszel, Cox proportional hazards) and multi-center, multi-investigator design provides structural counterweight.

Patients: Eligible patients were 18 to 85 years old with ECG-confirmed atrial fibrillation or atrial flutter persisting for 2 to 26 weeks. This duration window is notable: it excludes both very recent-onset atrial fibrillaation (where spontaneous conversion is more likely, potentially inflating apparent drug efficacy) and truly long-standing persistent or permanent atrial fibrillation (where pharmacological cardioversion is least likely to succeed), enriching for a population in an intermediate efficacy zone. The exclusion criteria were extensive and reflect appropriate caution for a class III antiarrhythmic: women of childbearing potential, recent myocardial infarction or unstable angina, recent cardiac surgery or angioplasty, significant sinus node disease, QRS >180 ms, baseline QT >440 ms, bradycardia <50 bpm, hypotension, and a long list of QT-interacting concomitant medications including other antiarrhythmics, tricyclics, and antihistamines. Patients with calculated creatinine clearance below 20 mL/min were excluded outright, with further dose stratification by renal function building on the lesson learned mid-course in DIAMOND-CHF.

Baseline Characteristics: A total of 325 patients across 37 centers were randomized to dofetilide 125 µg, 250 µg, or 500 µg twice daily, or placebo, with reasonably even distribution across the four arms (82, 82, 77, and 84 patients respectively). Mean age was 66–68 years across groups, with a strong male predominance (roughly 83–87% male). Atrial fibrillation was the primary diagnosis in 80–91% of patients across arms, with atrial flutter accounting for the remainder - though the proportion of flutter patients varied meaningfully by group (15% in the 125 µg arm vs. 9% in the 250 µg arm vs. 16% in the 500 µg arm vs. 20% in placebo), an imbalance the authors later note complicated interpretation of the dose-response relationship. The majority of patients (72%) were NYHA class II or III. Structural heart disease prevalence varied somewhat by arm (57% to 77%), and hypertensive disease was present in 38–60%. Anticoagulant use at entry was high and consistent across arms (88–91%), as was cardiac glycoside use (77–87%). These baseline imbalances, while not enormous, are the kind that can subtly influence a dose-ranging trial’s apparent dose-response gradient, and the authors explicitly flag the flutter/fibrillation imbalance as a contributing factor to an inconsistency in the 125 µg vs. 250 µg comparison.

Structural heart disease, defined broadly to include prior myocardial infarction, heart failure, ischemic or valvular disease, and cardiomyopathy was present in more than half of all patients, indicating this was not a population free of cardiac comorbidity despite the exclusion of recent acute events.

Procedures: The trial proceeded in two phases: an in-hospital conversion phase of three days or five doses, followed by a 12-month outpatient maintenance phase. All patients were admitted to telemetry for a minimum of three days regardless of treatment arm. Patients not converting pharmacologically after five doses underwent electrical cardioversion. After the first 105 patients were enrolled, the protocol was amended to add creatinine clearance-based dose adjustment, mirroring the same mid-trial correction applied in DIAMOND-CHF and reflecting an emerging, sponsor-wide recognition that fixed dosing was inadequate for this drug. The dosing algorithm additionally halved the dose for any QTc increase exceeding 15% from baseline, with withdrawal mandated for QT/QTc exceeding 550 ms or a 25% increase from baseline. Treatment comparisons were conducted by randomized group regardless of whether a dose was subsequently adjusted, an intention-to-treat approach to the dosing algorithm that is methodologically sound but means the “500 µg” arm in the results actually reflects a mixture of dosages in the 48% of patients who required downward adjustment. Anticoagulation was mandated before conversion and for a minimum of three to four weeks afterward, consistent with the thromboembolic risk of cardioversion, though continuation thereafter was left to local practice rather than protocol-mandated.

Endpoints: The primary efficacy endpoints were pharmacological conversion to sinus rhythm during the in-hospital phase and time to first atrial fibrillation/flutter relapse during the 12-month maintenance phase. Conversion success required maintenance of sinus rhythm for a minimum of 24 hours. The maintenance phase was analyzed in two separate populations: an intention-to-treat population in which patients who failed to convert were counted as relapsing at time zero, and a per-protocol-like population restricted to the 250 patients who successfully converted (either pharmacologically or electrically) and actually entered the maintenance phase. This dual-population approach is appropriate and transparent, but the headline maintenance figures reported in the abstract and widely cited thereafter - 40%, 37%, 58%, and 25% - come from the converted-only population, which is a post-randomization selected subgroup; readers should bear in mind that the intention-to-treat maintenance probabilities were roughly 20 percentage points lower across all arms. No formal prospective sample size calculation for a specific effect size is reported in the text, which is a gap for a trial explicitly framed as dose-ranging - the n of 325 appears to reflect practical recruitment targets rather than a power calculation tied to a pre-specified minimum clinically important difference in conversion or maintenance rates.

Results: Pharmacological cardioversion occurred in a clearly dose-dependent fashion: 6.1% with 125 µg, 9.8% with 250 µg, 29.9% with 500 µg, and 1.2% with placebo. The 250 µg and 500 µg doses both achieved statistical significance over placebo (P=0.015 and P<0.001 respectively), while 125 µg did not (P=0.098). Of those who converted pharmacologically, 70% did so within 24 hours and 91% within 36 hours - a clinically useful piece of information suggesting that failure to convert by 36 hours should prompt consideration of electrical cardioversion rather than continued waiting.

The maintenance-phase results showed dofetilide outperforming placebo in a dose-dependent manner among the 250 patients who successfully converted: probability of remaining in sinus rhythm at 12 months was 40% (125 µg), 37% (250 µg), 58% (500 µg), and 25% (placebo), with the 500 µg vs. placebo comparison reaching significance (P=0.001, hazard ratio 0.44, 95% CI 0.26–0.73). The 125 µg and 250 µg arms did not reach significance versus placebo (P=0.208 and P=0.104), and the inversion - 125 µg numerically outperforming 250 µg - is exactly the kind of irregularity the baseline flutter/fibrillation imbalance may explain, since flutter patients maintained sinus rhythm substantially better than fibrillation patients in covariate analysis (P= 0.0004), and the 250 µg arm had disproportionately fewer flutter patients (9% vs. 15–20% in other arms). Median time to atrial fibrillation relapse was 27 days on placebo versus 31, 179, and greater than 365 days for the 125, 250, and 500 µg dofetilide groups respectively - figures that, taken at face value, suggest a substantial and clinically meaningful delay in recurrence even where the binary 12-month maintenance comparison did not reach conventional significance.

The safety data, while reassuring, deserve close attention given dofetilide’s mechanism. Two cases of torsade de pointes occurred (0.8% of all patients on active drug), both within the first three days of therapy - one degenerating to ventricular fibrillation requiring defibrillation, the other similarly requiring electrical defibrillation and magnesium infusion. One sudden cardiac death, classified as proarrhythmic, occurred on day 8 in a patient receiving 500 µg (0.4% of all active-drug patients). Both torsade cases occurred in patients with calculated creatinine clearance above 60 mL/min - that is, in patients who were not renally impaired and would not have triggered dose reduction on that basis alone, indicating that renal-function-based dosing, while clearly risk-reducing, does not fully eliminate proarrhythmic risk even in patients with preserved renal function. Eleven patients were withdrawn for QT/QTc prolongation, seven of these within the first three days, reinforcing that the early monitoring window captures the great majority of clinically meaningful electrophysiological risk.

Conclusions: SAFIRE-D demonstrates that dofetilide produces dose-dependent pharmacological cardioversion of atrial fibrillation and atrial flutter, with the 500 µg twice-daily dose achieving a clinically meaningful 30% conversion rate against a 1.2% placebo rate, and that successful converters maintain sinus rhythm substantially longer on dofetilide than placebo at the highest dose tested. These are genuine, statistically supported efficacy findings, and the trial’s design, separating conversion and maintenance as distinct phases with distinct populations, allows a clear-eyed view of where the drug’s benefit actually accrues.

Readers should also note that the widely quoted maintenance percentages (40/37/58/25%) describe only the subset of patients who successfully converted to sinus rhythm in the first place, a population already enriched for treatment response, and that the intention-to-treat maintenance figures, which better reflect real-world expectations for an unselected patient starting dofetilide, were approximately 20 percentage points lower across all arms.

The safety profile, while favorable in aggregate, is not without weight. A proarrhythmic event rate of roughly 1% (two torsade cases plus one sudden death among 243 active-drug patients) in a population already enriched for low-risk QT and renal parameters by trial exclusion criteria is not a trivial figure to extrapolate to a broader, less rigorously screened population. Combined with the DIAMOND-CHF mortality-neutrality data, SAFIRE-D supports the case that dofetilide is a reasonable option for highly symptomatic atrial fibrillation or flutter of more than one week’s duration, provided it is initiated in a monitored inpatient setting with renal-function-based dosing and protocol-driven QTc surveillance.

Background: By the late 1990s, atrial fibrillation had been well-established as both a common complication of heart failure and an independent driver of clinical deterioration, yet the antiarrhythmic armamentarium for this population was meagre and dangerous. Class I agents had been discredited following CAST, and quinidine carried a threefold mortality signal. Amiodarone offered some efficacy but imposed a burden of extracardiac toxicity that limited its long-term use. Dofetilide, a selective IKr blocker without negative inotropic properties, emerged as a theoretically attractive candidate - rhythm control without the collateral damage of less selective agents. DIAMOND-CHF was designed to determine whether dofetilide affected mortality or morbidity in patients with symptomatic heart failure and reduced ejection fraction, with atrial fibrillation effects analyzed as a pre-specified substudy. The trial was funded entirely by Pfizer, the drug’s manufacturer; a Pfizer employee sat on the steering committee and writing group, and the study was coordinated by a Pfizer-contracted logistics firm. This relationship is worth bearing in mind, though the independent data and safety monitoring board and blinded events committee provide some structural mitigation.

Patients: Eligibility required symptomatic congestive heart failure with NYHA class III or IV symptoms in the preceding month, and a wall-motion index ≤1.2 on centrally reviewed echocardiography, corresponding roughly to an ejection fraction ≤35%. Patients were enrolled consecutively from 34 Danish hospitals during hospitalizations for new or worsening heart failure, an approach that enriched for high-risk, acutely decompensated patients rather than stable outpatients. A recent MI within seven days was an exclusion, as were significant conduction abnormalities untreated by pacing, corrected QT >460 ms, creatinine clearance <20 ml/min (stage 4 CKD or eGFR <30), and prior proarrhythmia - exclusions that were clinically sensible.

Baseline Characteristics: A total of 5,548 patients were screened, 2,531 met echocardiographic criteria, and only 1,518 (27% of those screened) were ultimately randomized. The two groups were well-matched across the major clinical variables. Mean age was 70 years in both arms, with a male predominance of approximately 72–75%. Ischemic etiology was present in roughly 67% of each group, prior myocardial infarction in approximately 51%, and diabetes in 19–20%. Median wall-motion index was 0.9 in both arms. Atrial fibrillation was present at baseline in 25% of the dofetilide group and 27% of the placebo group. Creatinine clearance averaged 57 ml/min in each arm. NYHA class distribution was similar, with the majority in class III (56% vs. 51%). The most notable population-level feature is the very low beta-blocker use (approximately 10% in both arms), a reflection of when this trial was conducted.

Procedures: All patients were admitted to hospital and underwent continuous cardiac monitoring for a mandatory 72-hour initiation period, with dose adjustments made according to QTc prolongation and renal function. This in-hospital loading protocol was not merely procedural, it was a direct safety intervention arising from recognition of dofetilide’s proarrhythmic risk window, with 76% of all torsade de pointes events occurring within the first three days. Notably, the dosing algorithm was modified mid-trial after 288 patients had been enrolled: the initial fixed-dose strategy was replaced by renal function-guided dosing based on emerging data from other dofetilide studies. This amendment introduces some heterogeneity - the 146 patients enrolled under the original protocol had a higher torsade de pointes rate (4.8%) than those enrolled after the amendment (2.9%) - and the authors separately analyzed mortality excluding the pre-amendment cohort. The study continued without interruption across four pre-planned interim analyses, all of which recommended continuation.

Endpoints: The primary endpoint was all-cause mortality, a commendably hard outcome in a population where the stakes were well-defined. Secondary endpoints included cardiovascular mortality, arrhythmic death, the composite of cardiac death or successful resuscitation after cardiac arrest, worsening heart failure hospitalization, and myocardial infarction. A pre-specified substudy examined conversion to and maintenance of sinus rhythm in the atrial fibrillation subgroup. The hospitalization endpoint for worsening heart failure required both admission and an escalation in heart failure medications, a composite definition that is clinically meaningful but introduces some subjectivity in medication adjustment decisions made at the site level. No formal hierarchy of secondary endpoints was pre-specified for multiplicity control, which matters when one secondary endpoint achieves significance and others do not. The atrial fibrillation substudy was pre-planned rather than post hoc, which strengthens its interpretive weight.

Results: The primary endpoint was neutral: 311 deaths in the dofetilide group (41%) versus 317 in placebo (42%), hazard ratio 0.95 (95% CI 0.81–1.11). The Kaplan-Meier curves run nearly parallel throughout follow-up without early divergence suggesting transient benefit or harm. Subgroup analyses across 14 pre-defined variables, including the atrial fibrillation subgroup (HR 1.00, 95% CI 0.74–1.35) and the non-AF subgroup (HR 0.94, 95% CI 0.78–1.13), showed no signal of heterogeneity in either direction.

The hospitalization endpoint for worsening heart failure was positive: 30% in the dofetilide group versus 38% in placebo, risk ratio 0.75 (95% CI 0.63–0.89), an absolute risk reduction of approximately 8 percentage points and a number needed to treat of roughly 12–13 over 18 months. The effect was present in both atrial fibrillation and non-atrial fibrillation subgroups. The authors speculate that rhythm control in the atrial fibrillation subgroup may have contributed, but the benefit in patients without baseline atrial fibrillation, where no such mechanism applies, leaves the biological pathway unexplained. The definition requiring both hospitalization and medication escalation adds some rigor, but center-level variability in hospitalization thresholds across 34 sites in 1990s Denmark cannot be excluded. Without a pre-specified multiplicity hierarchy this finding should be treated as hypothesis-generating.

Figure below taken directly from the NEJM publication.

In the atrial fibrillation sub-study, dofetilide demonstrated clear pharmacologic efficacy: spontaneous cardioversion to sinus rhythm at one month occurred in 12% of the dofetilide group versus 1% of placebo, and at 12 months the pharmacologic cardioversion rate was 44% versus 13%. Among patients successfully cardioverted, dofetilide was highly effective at maintaining sinus rhythm (HR for atrial fibrilltion recurrence 0.35, 95% CI 0.22–0.57). Despite this rhythm benefit, the composite of death, stroke, and systemic embolism in the atrial fibrillation subgroup was identical between arms (52/190 vs. 54/201, P=0.85) - a finding that tempers enthusiasm for the rhythm control signal and anticipates what AFFIRM would later formalize. A pooled DIAMOND substudy subsequently reported that patients who achieved and maintained sinus rhythm had substantially lower mortality (RR 0.44, 95% CI 0.30–0.64), but this is a heavily confounded post-hoc observation: patients who convert and stay in sinus rhythm are inherently lower risk, and this association cannot be treated as evidence of a treatment effect.

The safety data deserve prominent attention. Torsade de pointes was observed in 25 of 762 DIAMOND-CHF patients (3.3%), compared with 0.9% in DIAMOND-MI and 0.8–2.5% across other dofetilide trial populations. Two of the 25 episodes were fatal and 15 required electrical cardioversion. Risk factors for torsade de pointes included higher dose, female sex, baseline QT >450 ms, greater QTc increase during loading, and a history of sustained ventricular tachycardia - and it is important to note that patients with QTc >460 ms, resting heart rate <50 bpm, or prior polymorphic ventricular tachycardia were excluded from the trial.

Conclusions: DIAMOND-CHF answers its primary question cleanly: dofetilide does not reduce all-cause mortality in patients with heart failure and reduced ejection fraction. The result is not a borderline negative, the hazard ratio of 0.95 with a narrow confidence interval provides genuine reassurance that the drug does not actively harm survival at the population level, distinguishing it from class I agents and d-sotalol. This mortality neutrality, combined with the rhythm control efficacy in the atrial fibrillation subgroup, formed the evidentiary basis for FDA approval in 1999 and has sustained dofetilide’s niche clinical role since.

That niche is geographically narrow. Dofetilide is marketed exclusively in the United States; it is not available in Europe or Australia. After initially receiving marketing authorisation in Europe in 1999, Pfizer voluntarily withdrew it in 2004, citing commercial rather than safety reasons. In Canada, dofetilide is not approved by Health Canada and is accessible only through the Special Access Program for patients with serious or life-threatening conditions in whom conventional therapies have failed. The drug is effectively a US-only agent, and the regulatory divergence likely reflects not just commercial calculus but a judgment that the proarrhythmic burden relative to available alternatives did not justify routine approval elsewhere.

The reduction in heart failure hospitalizations is statistically significant, but the biological mechanism is unexplained, the endpoint definition is susceptible to site-level variability, and the absence of a pre-specified multiplicity hierarchy means it cannot be elevated to a primary conclusion. Furthermore, any reduction in heart failure hospitalization is likely offset by an increase in non-heart failure hospitalization, which is not reported in the text. The 3.3% rate of torsade de pointes (2.9% after protocol amendment) is not trivial and real-world patients may experience higher risk outside controlled trial conditions, where acutely decompensated patients, electrolyte disturbances, and co-prescribed QT-prolonging agents are likely more common and where the meticulous monitoring of the DIAMOND protocol is more difficult to replicate.

The key message from DIAMOND-CHF is not that dofetilide is beneficial for patients with atrial fibrillation. The atrial fibrillation findings were derived from a subgroup analysis and did not demonstrate a mortality benefit. Rather, the principal conclusion of the trial is that dofetilide does not increase mortality in patients with systolic heart failure and can be used when a rhythm-control strategy is indicated.

Background: For two decades the rate-versus-rhythm question in atrial fibrillation appeared settled in the negative. AFFIRM, RACE, STAF, and the heart-failure trial of Roy and colleagues had each failed to show that pursuing sinus rhythm with antiarrhythmic drugs improved outcomes over controlling the ventricular rate and anticoagulating. The conventional reading was that rhythm control conferred no prognostic advantage and that the drug toxicity it entailed offset whatever benefit sinus rhythm itself might provide. EAST-AFNET 4 set out to challenge that conclusion by changing several variables at once. The investigators argued that the earlier trials had enrolled patients with established, long-standing atrial fibrillation, that they largely predated catheter ablation, and that background anticoagulation and cardiovascular therapy had been less complete. Their hypothesis was that initiating rhythm control early, within a year of diagnosis, in patients with cardiovascular comorbidity, and permitting ablation as part of the strategy would reduce hard cardiovascular events. It is worth noting at the outset that this design bundles at least three distinct changes (early timing, ablation availability, and a guideline-concordant treated background) into a single strategy comparison.

Patients: Eligible patients had early atrial fibrillation, defined as diagnosis no more than 12 months before enrollment, plus cardiovascular risk: age over 75, prior stroke or TIA, or two of a list including age over 65, female sex, heart failure, hypertension, diabetes, severe coronary disease, chronic kidney disease, or left ventricular hypertrophy. The median time since diagnosis was 36 days, so this was a genuinely early-AF population, distinct from the AFFIRM and RACE cohorts. The trial’s own discussion concedes the key selection filter: all enrolled patients had to be deemed eligible for either rate or rhythm control, which the authors acknowledge probably excluded the most symptomatic patients - those for whom rhythm control is least optional.

Baseline Characteristics: The trial enrolled 2,789 patients across 135 sites in 11 European countries and randomized them 1:1 to early rhythm control (n=1,395) or usual care (n=1,394). The mean age was 70 years and 46% were women, with a mean body-mass index of 29. Atrial fibrillation was a first episode in 38%, paroxysmal in 36%, and persistent in 26%. Notably, about 54% of patients were already in sinus rhythm at randomization. Symptom burden was modest: roughly 30% were asymptomatic (European Heart Rhythm Association - EHRA I), 52% EHRA II, and only about 18% EHRA III–IV. The mean CHA₂DS₂-VASc score was 3.4. Hypertension was present in 88%, stable heart failure in 28%, diabetes in 25%, severe coronary artery disease in 17%, and chronic kidney disease in 13%; the mean left atrial diameter was 44 mm. More than 90% were on oral anticoagulation at discharge. The groups were generally well matched, with the medication differences traceable to the allocated treatment: beta-blocker use was lower in the rhythm arm (76.2% vs. 85.5%) and digoxin less common (3.3% vs. 6.1%), both p<0.001 and both downstream of the rate-control mandate in usual care; statin use was slightly higher in the rhythm arm (45.2% vs. 40.8%; p=0.018). The adjusted analysis accounting for baseline covariates was essentially unchanged from the primary result.

Procedures: The trial was investigator-initiated, open-label, with blinded outcome adjudication (a PROBE design). Randomization was 1:1, stratified by site with variable block lengths for concealment. All serious adverse events were adjudicated by an endpoint review committee unaware of treatment assignment, which limits ascertainment bias for hard endpoints but cannot eliminate it for behavior-sensitive outcomes such as hospitalization.

In the early-rhythm-control arm, 94.8% received an antiarrhythmic drug or ablation. The strategy was predominantly pharmacological: most patients began on antiarrhythmic drugs, flecainide most commonly, and at two years, of the 65.1% still on active rhythm control, only 270 of 1,395 (about 19%) had undergone ablation versus 638 on drugs. This is worth flagging because the investigators lean on ablation to explain the trial’s success relative to prior drug-based trials, yet ablation was the minority intervention. Usual care was not a do-nothing arm: it mandated guideline-based anticoagulation, rate control, and treatment of cardiovascular conditions, and permitted rhythm control for symptom relief. At two years, 14.6% of usual-care patients were nonetheless receiving rhythm-control therapy, and only 65% of the rhythm arm remained on therapy — so the two strategies were not cleanly separated, which biases a strategy comparison toward the null. Sites were structured as ablation centers (”A sites”) with nested referring “D sites,” a clustering handled in the statistical models and relevant to the withdrawal analysis below.

Figure below is taken directly from the published article in NEJM.

The trial was primarily publicly funded (German Ministry of Education and Research, the German Center for Cardiovascular Research, the EU, the British Heart Foundation, and the Leducq Foundation), with additional industry support from St. Jude Medical–Abbott and Sanofi, both of which held non-voting seats on the executive steering committee.

Endpoints: The first primary outcome was a composite of cardiovascular death, stroke, or hospitalization for worsening heart failure or acute coronary syndrome. The second primary outcome was the number of nights spent in hospital per year, a soft count-type endpoint that directly addressed the historical concern that rhythm control may increase hospitalization. Sinus-rhythm maintenance, left ventricular function, and symptom scores were treated as secondary rather than as surrogates for clinical benefit, which is to the trial’s credit. The analysis followed a group-sequential design with three interim analyses and O’Brien–Fleming stopping boundaries; competing events (non-cardiovascular death, withdrawal) were handled with Aalen–Johansen cumulative-incidence methods rather than Kaplan–Meier.

The objective was to show the superiority of early rhythm control over usual care.

This study used an event-driven design, with each of the two primary endpoints evaluated separately when comparing the treatment groups. To maintain a total two-sided type I error rate of 5%, significance levels were allocated as 4% for the first primary endpoint and 1% for the second. A reduction of 20% in the yearly incidence of the first primary endpoint was considered clinically meaningful. Power calculations indicated that approximately 685 outcome events would be required to detect such a difference with 80% statistical power.

Results: At 24-months, 60.1% were in sinus rhythm in the usual care arm compared to 80.7% in the rhythm control arm.

The trial was stopped early for efficacy at the third interim analysis after a median 5.1 years of follow-up. The first primary outcome occurred in 249 early-rhythm-control patients (3.9 per 100 person-years) versus 316 usual-care patients (5.0 per 100 person-years), with a hazard ratio of 0.79 (96% CI 0.66–0.94; p=0.005). The relative reduction of 21% corresponds to an absolute difference of 1.1 events per 100 person-years — a number-needed-to-treat in the high teens to low twenties over five years, a real but modest absolute benefit.

Decomposition of the composite is reassuring. Cardiovascular death was 0.72 (95% CI 0.52–0.98), stroke 0.65 (0.44–0.97), heart-failure hospitalization 0.81 (0.65–1.02), and acute coronary syndrome hospitalization 0.83 (0.58–1.19). The components move concordantly, and, unusually, the two that reached nominal significance were the hard endpoints rather than the soft hospitalization components, the opposite of the usual pattern in which composites are carried by their softest elements. These component confidence intervals are unadjusted for multiplicity, and the trial explicitly cautions against using them to infer definitive effects, so the death and stroke results are supportive rather than standalone proof.

The second primary outcome was null: 5.8 versus 5.1 nights per year (p=0.23). The supplement shows the investigators initially modeled two opposing non-significant trends - possibly fewer hospitalizations but longer stays once admitted - before simplifying to a model yielding an incidence rate ratio of 1.08 (p=0.226), with sensitivity analyses (Fisher–Pitman p=0.808) confirming the null. The honest reading is that early rhythm control did not increase hospital nights, not that it reduced them.

The cumulative-incidence curves separated early and diverged gradually across follow-up, consistent with a steadily accumulating true effect rather than an early artifact; the at-risk numbers thin to the dozens after year six, so the right-hand portion should not be over-read. The most important issue the supplement surfaces is a significant excess of withdrawals in the early-rhythm-control arm. Differential withdrawal can bias a time-to-event analysis, and this was not anticipated in the statistical analysis plan. The investigators added a post-hoc analysis, found the excess concentrated in ablation-only A sites, and re-estimated the primary outcome with withdrawal as an additional competing event: the sub-distribution hazard ratio was 0.76 (95% CI 0.65–0.90; p=0.002), essentially unchanged. The convergence of the primary, covariate-adjusted (0.78), Gray’s-test (p=0.003), and withdrawal-corrected (0.76) estimates is the strongest internal evidence of robustness, though a post-hoc correction for unanticipated differential dropout is reassurance rather than the protection of no imbalance at all.

The treatment effect was broadly consistent across predefined subgroups, including by symptom status, obesity, heart failure, and center type. The forest plot shows wide, overlapping intervals for the small strata, so directional consistency is the appropriate takeaway rather than any individual subgroup estimate. One methodological caveat applies throughout the secondary analyses: 58.6% of patients had at least one missing value across the variables and timepoints used, handled by multiple imputation with 60 repetitions - a burden that matters little for the adjudicated hard endpoints but considerably for the continuous secondary outcomes.

Conclusions: In patients with recently diagnosed atrial fibrillation and cardiovascular risk factors, a strategy of early rhythm control reduced a composite of cardiovascular events compared with a strong, contemporary usual-care comparator. The trial is well conducted and largely independently funded, with several features that earn it credibility. Several aspects deserve careful consideration.

First, looking at the flow diagram above: the trial reports 2,810 patients assessed for eligibility, of whom only 21 were excluded, with the rest randomized. This is not realistic. Atrial fibrillation is a very common condition in the elderly, and it took investigators roughly 4.5 years and 135 centers to recruit this cohort. Over that time span, across that many centers, they likely encountered tens of thousands of AF patients or more. The true number of patients assessed for eligibility is almost certainly far higher than what's shown in this diagram - what's reported looks more like the number who cleared eligibility screening than the number actually screened. Without knowing the screened-to-enrolled ratio, it's very difficult to judge how selected this population was and who these results actually apply to. This is a major limitation.

Second, the result is a strategy effect, not a verdict on any single modality. Early timing, ablation availability, and a comprehensively treated background were changed together, and the trial cannot apportion the benefit among them. The investigators’ emphasis on ablation is particularly hard to sustain given that ablation was the minority intervention; most of the rhythm-control benefit, if causal, was delivered by antiarrhythmic drugs on a well-managed background. Furthermore, at 24-months, 60.1% were in sinus rhythm in the usual care arm compared to 80.7% in the rhythm control arm. Can this 20% difference be responsible for clinical benefit seen?

Third, the open-label design introduces performance design.

Forth, the magnitude and certainty of the effect should be read conservatively. The absolute benefit is about 1.1 events per 100 person-years, and the trial was stopped early at an interim boundary - a procedure that biases the point estimate upward and is consistent with a true effect nearer the 0.94 upper bound. The concordance of the hard components with the composite, however, makes the qualitative finding more durable than the precise hazard ratio.

Fifth, the population defines the boundary of generalization. These were early, often minimally symptomatic patients, more than half already in sinus rhythm at randomization, explicitly screened to exclude the most symptomatic. The result does not extend cleanly to late, long-standing, or unselected atrial fibrillation, nor to the highly symptomatic patients for whom rhythm control is pursued for symptom relief regardless of prognosis.

RACE and AFFIRM showed that chasing sinus rhythm with drugs in established atrial fibrillation produced some toxicity without prognostic gain, EAST-AFNET 4 shifts the prior for early disease, suggesting that a comprehensive early strategy on a guideline-concordant background can reduce hard events. What it does not do is identify the active ingredient. The clinical reading that holds up best is not “treat every new atrial fibrillation patient with rhythm control” - it is “early, comprehensive management in comorbid patients may improve outcomes, the active component is unresolved, and the result should not be stretched to populations the trial did not enroll.” The early intervention hypothesis is biologically compelling. The better drug and ablation hypothesis is mechanistically sound. But the structured follow-up and open-label bias explanations cannot be dismissed, and the replication of this finding in a blinded or pragmatic real-world setting remains the unmet need that would definitively settle the question.

Background: Dronedarone had been approved on the basis of the ATHENA trial, which showed that the drug reduced the composite endpoint of cardiovascular hospitalization or death in patients with paroxysmal or persistent atrial fibrillation. The ATHENA result drove enthusiastic adoption of dronedarone in much of the developed world. A natural question followed: if dronedarone is beneficial in patients with intermittent atrial fibrillation, what about patients with permanent atrial fibrillation? These patients are older, sicker, and at higher absolute risk of the very cardiovascular events that ATHENA appeared to reduce. The drug had several mechanisms - heart-rate slowing, blood-pressure lowering, antiadrenergic effects, ventricular antiarrhythmic effects in animal models - that might plausibly benefit this population even without the rhythm-control component. The authors hypothesized that dronedarone would reduce major vascular events in high-risk patients with permanent atrial fibrillation.

This hypothesis required a leap. ANDROMEDA, an earlier trial of dronedarone in patients with advanced symptomatic heart failure, had been terminated prematurely for excess mortality in the dronedarone arm. ATHENA had excluded such patients. The question of whether dronedarone was beneficial, neutral, or harmful in patients between the ATHENA and ANDROMEDA populations - older, with more advanced cardiovascular disease, but without decompensated heart failure - was genuinely unsettled. PALLAS sought to answer it.

Patients: Eligible patients were at least 65 years of age with permanent atrial fibrillation or flutter (documented on ECG both within 14 days before randomization and at least 6 months previously, with no intervening sinus rhythm and no plan to restore sinus rhythm). At least one additional risk factor was required: coronary artery disease, previous stroke or TIA, symptomatic heart failure (NYHA class II or III with hospitalization for heart failure in the previous year but not the most recent month), LVEF ≤40%, peripheral arterial disease, or the combination of age ≥75 with hypertension and diabetes. Major exclusion criteria were paroxysmal or persistent (i.e., non-permanent) atrial fibrillation, implantable cardioverter-defibrillator, sustained daytime bradycardia <50 bpm, or QTc >500 msec (>530 msec if paced).

Baseline Characteristics: The trial enrolled 3,236 patients across 489 sites in 37 countries and randomized them 1:1 to dronedarone 400 mg twice daily (n=1,619) or placebo (n=1,617) before being terminated early. The mean age was 75 years, 64.6% were male, and 69% had had permanent atrial fibrillation for more than 2 years. Coronary artery disease was present in 41%, prior stroke or TIA in 27.6%, LVEF ≤40% in 21%, and peripheral arterial disease in 12.4%. The mean CHADS₂ score was 2.85, and 88.7% had a CHADS₂ score ≥2. The most striking feature of this cohort relative to ATHENA was the burden of heart failure: about two-thirds had a history of heart failure (NYHA class I in 13.7%, class II in 45.8%, class III in 8.2%), and 14.6% met the inclusion criterion for symptomatic heart failure with recent hospitalization. Background therapy was guideline-concordant: 74% on beta-blockers, 84% on vitamin K antagonists, 33.5% on digoxin, 70% on diuretics, 76.4% on ACE inhibitor or ARB, and 57.4% on statins. Baseline characteristics were well balanced between the two groups.

Procedures: The trial was double-blind, placebo-controlled, with randomization stratified by center. Patients were seen on days 7 and 30, at 4 months, and every 4 months thereafter. Investigators were advised to use digoxin with caution because of a known P-glycoprotein interaction, and serum digoxin was measured on day 7. Drugs known to prolong the QT interval were prohibited. Liver function tests were initially performed at each office visit; after a protocol amendment in January 2011 (responding to reports of dronedarone-associated hepatotoxicity in post-marketing surveillance), monitoring was intensified to monthly for the first 6 months and bimonthly thereafter. The trial was sponsored by Sanofi-Aventis. Operations and steering committees designed the study. Data collection and analysis were performed by the Population Health Research Institute in Hamilton, Ontario - an academic statistical coordinating center independent of the sponsor. This is a structurally stronger arrangement than ATHENA, in which the sponsor performed both data collection and analysis.

The trial was planned to enroll 10,800 patients over 2 years with 1 year of additional follow-up, providing 90% power to detect a 20% relative reduction in the first coprimary outcome (assuming a 4.5% one-year placebo event rate). One efficacy interim analysis was planned at 50% of expected events using a modified Haybittle-Peto boundary. The data monitoring committee was charged with recommending termination only for “clear, consistent, and persistent evidence of net harm that overwhelmed benefit.” On July 5, 2011, after enrollment of 3,236 patients with a median follow-up of only 3.5 months, the data monitoring committee recommended termination for safety reasons.

Endpoints: PALLAS used two coprimary endpoints. The first was a composite of stroke, myocardial infarction, systemic embolism, or death from cardiovascular causes. The second was unplanned cardiovascular hospitalization or death from any cause, mirroring the ATHENA primary endpoint. Other prespecified outcomes included death from cardiovascular causes, death from arrhythmia, recurrent cardiovascular hospitalization, total cardiovascular hospital nights, acute coronary syndrome, stroke or systemic embolism, hospitalization for heart failure or heart-failure episode, and death from any cause. All outcomes except heart-failure episodes without hospitalization were adjudicated by an independent committee blinded to study-group assignments.

Results: The first coprimary outcome occurred in 43 patients receiving dronedarone and 19 patients receiving placebo (hazard ratio 2.29; 95% CI 1.34–3.94; p=0.002) at a median follow-up of 3.5 months. The annualized event rate was 8.2 per 100 patient-years in the dronedarone group versus 3.6 in the placebo group. The Kaplan-Meier curves diverged within the first month and continued to separate throughout follow-up. The second coprimary outcome occurred in 127 dronedarone patients and 67 placebo patients (HR 1.95; 95% CI 1.45–2.62; p<0.001).

The decomposition of these composites is what makes PALLAS clinically definitive rather than statistically suggestive. Death from any cause occurred in 25 dronedarone patients versus 13 placebo patients (HR 1.94; 95% CI 0.99–3.79; p=0.049). Cardiovascular death occurred in 21 versus 10 (HR 2.11; 95% CI 1.00–4.49; p=0.046). Arrhythmic death occurred in 13 versus 4 (HR 3.26; 95% CI 1.06–10.00; p=0.03). Stroke occurred in 23 versus 10 (HR 2.32; 95% CI 1.11–4.88; p=0.02). Hospitalization for heart failure occurred in 43 versus 24 (HR 1.81; 95% CI 1.10–2.99; p=0.02). Heart-failure episode or hospitalization combined occurred in 115 versus 55 (HR 2.16; 95% CI 1.57–2.98; p<0.001). Every signal moves in the same direction and most reach statistical significance despite the trial having accrued only a fraction of its planned follow-up.

The physiologic effects of dronedarone behaved as expected. At one month, heart rate was reduced by 7.6 bpm in the dronedarone arm versus an increase of 0.1 bpm in placebo. Systolic blood pressure fell by 3.5 mmHg versus 1.7 mmHg. QTc increased by 8 msec versus 2 msec. None of these intermediate measures translated into the clinical benefit the authors hypothesized. Serum digoxin concentration on day 7 was 33% higher in dronedarone patients (1.2 vs. 0.9 ng/ml; p<0.001), confirming the P-glycoprotein interaction in vivo. Time in therapeutic INR range was modestly lower in the dronedarone arm (55.6% vs. 58.6%; p=0.02), but the magnitude is too small to explain the more than doubling in stroke risk. Premature drug discontinuation was 21% vs. 11% (p<0.001) over a median 3.5 months - an extraordinary differential given how brief the follow-up was. Subgroup analyses showed remarkably consistent harm across age, atrial fibrillation duration, left ventricular ejection fraction strata, NYHA class, CHADS₂ score, heart rate, blood pressure, digoxin use, beta-blocker use, and region. The point estimates for the first coprimary outcome ranged from 1.54 to 5.43; none favored dronedarone.

Conclusions: PALLAS is one of the cleaner examples in cardiology of a trial that definitively answers the question it was designed to ask, and answers it in the direction opposite to the hypothesis. In patients with permanent atrial fibrillation and additional cardiovascular risk factors, dronedarone approximately doubled the rate of stroke, cardiovascular death, and heart-failure hospitalization. The signal was detected within months, the data monitoring committee terminated the trial appropriately, and the conclusion that dronedarone should not be used in this population, has been adopted into guidelines and regulatory labeling worldwide.

Several aspects of this trial deserve emphasis. First, the magnitude and consistency of harm are striking. Every prespecified clinical endpoint moved in the same direction. The point estimate for cardiovascular death was 2.11, for stroke 2.32, for arrhythmic death 3.26, and for heart-failure hospitalization 1.81. The subgroup analyses showed no subset of patients in whom dronedarone was beneficial or even neutral. This is not the pattern of a chance finding driven by a single endpoint or a single subgroup; it is the pattern of a drug that is harmful across the trial population.

Second, the early termination merits comment as a methodological matter. The authors acknowledge that early termination reduces statistical power and may bias point estimates away from the null. This is technically correct, and a longer trial might have shown somewhat smaller hazard ratios. But the direction of the effect would not have changed, and the data monitoring committee’s decision to stop the trial was the right one - continuing to randomize patients to a drug that was doubling their stroke risk would have been indefensible. The early termination is not a weakness of the trial’s conclusion; it is evidence of how clear the signal was.

Third, PALLAS provides a useful retrospective lens on ATHENA. The trials enrolled overlapping but distinct populations. ATHENA enrolled younger patients with paroxysmal or persistent atrial fibrillation and less cardiovascular disease, and showed benefit. PALLAS enrolled older patients with permanent atrial fibrillation and more cardiovascular disease, and showed harm. ANDROMEDA enrolled patients with severe heart failure and recent decompensation, and showed even greater harm. The pattern across these three trials suggests that dronedarone’s benefit in ATHENA was driven largely or entirely by reducing atrial fibrillation hospitalizations through suppression of recurrence, and that in populations where this mechanism does not apply (because the patient is in permanent atrial fibrillation) or is overwhelmed by other effects (negative inotropy, proarrhythmia, etc), the drug is net harmful.

A framing the authors do not explicitly offer but we think is instructive: the central problem with ATHENA’s interpretation was treating “reduction in cardiovascular hospitalization” as if it were a fundamental cardiovascular benefit rather than a mechanically expected consequence of suppressing the arrhythmia that drives those hospitalizations. PALLAS removes the rhythm-control mechanism (these patients are in permanent atrial fibrillation) and reveals what dronedarone does in its absence - it increases stroke, heart failure, arrhythmic death, and cardiovascular mortality. The most plausible reading of these three trials together is that dronedarone’s apparent benefit in ATHENA was a narrow effect on atrial fibrillation-related hospitalization, while its toxicities - proarrhythmic, negative inotropic - were present all along but masked by that benefit in the population studied. ATHENA’s modest mortality reduction (HR 0.84; p=0.18 for all-cause, HR 0.71; p=0.03 for CV) was always more fragile than the trial’s framing acknowledged, and PALLAS suggests it may not survive in a different patient mix.

The clinical implication of PALLAS is unambiguous and has been adopted: dronedarone is contraindicated in patients with permanent atrial fibrillation. In our view, the broader implication for the atrial fibrillation literature is that any antiarrhythmic drug demonstrating a “cardiovascular outcomes” benefit driven primarily by reduced hospitalization for the arrhythmia it suppresses should be viewed with skepticism. The drug works mechanistically, the hospitalization endpoint moves, and the question is whether the patient is actually better off - a question that is not answered by the composite. PALLAS does not invalidate ATHENA, but it does considerably tighten the population in which dronedarone can be defended, and it should temper enthusiasm for similar pharmacotherapy claims in the absence of mortality benefit.

We commend the data monitoring committee for terminating the trial promptly and the academic statistical coordinating center for the credibility this independent oversight conferred on the result. A weaker oversight structure might have allowed continued enrollment in pursuit of statistical power. PALLAS demonstrates that the trial machinery, when functioning correctly, can identify drug harm rapidly and protect patients - both those in the trial and the much larger population who would otherwise have received the drug in routine practice.

Background: Atrial fibrillation is the most common sustained cardiac arrhythmia, with a U.S. prevalence approaching 1% overall and rising to 9% in patients over 80 years of age. Hospitalizations for atrial fibrillation roughly tripled in the two decades preceding ATHENA, creating a substantial public health and economic burden. The pharmacologic options for rhythm control had remained largely unchanged since the 1980s and were limited by modest efficacy, proarrhythmia, or extracardiac toxicity. Amiodarone, the most effective antiarrhythmic available, carries well-documented thyroid, pulmonary, hepatic, and dermatologic toxicities that limit long-term use. Prior trials including AFFIRM had shown that rhythm control as a strategy did not improve survival over rate control, and no antiarrhythmic had been shown to reduce mortality in patients with atrial fibrillation.

Dronedarone is a benzofuran derivative structurally similar to amiodarone but lacking the iodine moiety and with an added methane-sulfonyl group intended to reduce lipophilicity, shorten the half-life to approximately 24 hours, and reduce tissue accumulation. The drug had been shown to maintain sinus rhythm better than placebo in two prior trials. A separate trial (ANDROMEDA), in patients with advanced symptomatic congestive heart failure and recent decompensation, was terminated prematurely because of excess mortality in the dronedarone arm. The ATHENA trial was designed to determine whether dronedarone would reduce the composite endpoint of first cardiovascular hospitalization or death from any cause in patients with atrial fibrillation and additional risk factors.

Patients: Eligible patients had paroxysmal or persistent atrial fibrillation or flutter and at least one additional risk factor: age ≥70 years, hypertension on two or more agents, diabetes, prior stroke/TIA/systemic embolism, left atrial diameter ≥50 mm, or LVEF ≤40%. A 12-lead ECG documenting atrial fibrillation or flutter within 6 months of randomization was required, as was a second 12-lead ECG showing sinus rhythm within the same period. Patients could be enrolled while in sinus rhythm or while in atrial fibrillation (with planned cardioversion after appropriate anticoagulation).

The exclusion criteria are critical to interpreting this trial. Patients with permanent atrial fibrillation were excluded. Patients with decompensated heart failure within the previous 4 weeks, NYHA class IV congestive heart failure, bradycardia <50 bpm, PR interval >0.28 seconds, sinus node disease without pacemaker, or hemodynamic instability were excluded. eGFR <10 ml/min, potassium <3.5 mmol/L, and concurrent class I or III antiarrhythmics were also exclusions. During enrollment, the inclusion criteria were modified because overall mortality was lower than expected - patients <70 years became ineligible, while patients ≥75 were eligible without other risk factors. This protocol amendment, implemented on March 8, 2006, deliberately enriched the population to recover statistical power and is worth flagging because it occurred mid-trial.

Baseline Characteristics: The trial enrolled 4,628 patients across 551 centers in 37 countries and randomized them 1:1 to dronedarone 400 mg twice daily (n=2,301) or placebo (n=2,327). The mean age was 71.6 years and 46.9% of participants were female. Twenty-five percent of patients had atrial fibrillation or flutter at randomization. The predominant comorbidity was hypertension (86.3%), with coronary heart disease in 30.4%, valvular heart disease in 16.4%, and nonischemic cardiomyopathy in 5.5%. Structural heart disease was present in 59.6% of patients with available data. A history of NYHA class II or III heart failure was present in 21.2% (class II in 17.1%, class III in 4.4%). LVEF was <45% in 11.9% and <35% in 3.9%. Baseline medication included: beta-blockers in 70.6%, ACE inhibitors or ARBs in 69.5%, vitamin K antagonists in 60.2%, statins in 38.7%. The two groups were well balanced except that the dronedarone group had significantly more women (49.2% vs. 44.6%; p=0.002).

Procedures: The trial was double-blind and placebo-controlled, with randomization stratified by center and by the presence or absence of atrial fibrillation or flutter at randomization. The follow-up schedule called for clinical evaluations at days 7 and 14, then at months 1, 3, 6, 9, and 12, and every 3 months thereafter. Minimum follow-up was 12 months with a common end date of December 30, 2007. Mean follow-up was 21±5 months.

The trial was sponsored by Sanofi-Aventis. The protocol was designed by the steering committee in collaboration with the sponsor. Data collection, management, and statistical analysis were performed by the sponsor. The first draft of the manuscript was written by the principal investigator (an academic author), with subsequent revisions by all authors. Two of the seven listed authors were Sanofi-Aventis employees holding shares in the company. The remaining academic authors disclosed extensive consulting fees, research grants, and lecture fees from Sanofi-Aventis. This sponsorship structure is not unusual for industry trials of this era but bears on how the discussion section frames the results.

Endpoints: The primary endpoint was the first hospitalization due to cardiovascular events or death from any cause. The composite is heavily weighted toward hospitalization, which dominated the event count. Hospitalization due to atrial fibrillation specifically is an endpoint where blinding matters enormously: physicians who suspect a patient is on placebo may have a lower threshold to admit when atrial fibrillation recurs. Secondary endpoints were death from any cause, death from cardiovascular causes, and first cardiovascular hospitalization. Deaths were adjudicated in a blinded fashion using a modified Hinkle-Thaler classification. The trial was powered (80%) to detect a 15% reduction in the primary endpoint assuming a 20% one-year placebo event rate, requiring 970 primary events and approximately 4,300 patients.

Results: The primary endpoint occurred in 734 patients (31.9%) in the dronedarone group and 917 patients (39.4%) in the placebo group (hazard ratio 0.76; 95% CI 0.69–0.84; p<0.001). The absolute risk reduction was 7.5 percentage points over a mean follow-up of 21 months, yielding a number needed to treat of approximately 13. The Kaplan-Meier curves separated early and continued to diverge throughout follow-up. The subgroup analysis (not prespecified) showed consistent benefit across age, sex, atrial fibrillation status at randomization, structural heart disease, heart failure, left ventricular ejection fraction strata, and concomitant medication use.

Decomposing the primary endpoint reveals where the result actually lives. Of 734 primary events in the dronedarone group, 675 (92%) were cardiovascular hospitalizations and only 59 (8%) were deaths. The composite was driven almost entirely by hospitalizations. Within hospitalizations, the dominant driver was admission for atrial fibrillation itself: 335 (14.6%) in the dronedarone group vs. 510 (21.9%) in the placebo group (HR 0.63; p<0.001). Hospitalizations for heart failure, syncope, and ventricular arrhythmia or cardiac arrest did not differ significantly between groups. Hospitalization for acute coronary syndrome was reduced (62 vs. 89; HR 0.70; p=0.03).

The mortality results are where the trial gets interesting. Death from any cause occurred in 116 patients (5.0%) in the dronedarone group and 139 patients (6.0%) in the placebo group (HR 0.84; 95% CI 0.66–1.08; p=0.18). This is not a statistically significant difference. Death from cardiovascular causes was 63 (2.7%) vs. 90 (3.9%) (HR 0.71; 95% CI 0.51–0.98; p=0.03), driven primarily by a reduction in death from cardiac arrhythmia (26 vs. 48; HR 0.55; 95% CI 0.34–0.88; p=0.01). Death from non-arrhythmic cardiac causes, non-cardiac vascular causes, and non-cardiovascular causes did not differ significantly.

Premature discontinuation of study drug was substantial and similar between groups: 30.2% with dronedarone vs. 30.8% with placebo. However, discontinuation due to adverse events was significantly higher with dronedarone (12.7% vs. 8.1%). Bradycardia (3.5% vs. 1.2%), QT prolongation (1.7% vs. 0.6%), diarrhea (9.7% vs. 6.2%), nausea (5.3% vs. 3.1%), rash (3.4% vs. 2.0%), and serum creatinine elevation (4.7% vs. 1.3%) were all significantly more common with dronedarone. Pulmonary symptoms, interstitial lung disease, and thyroid abnormalities were not significantly increased — but the authors appropriately note that amiodarone-type pulmonary toxicity typically emerges after 2 years of therapy, and mean follow-up here was only 21 months.

Conclusions: In patients with atrial fibrillation and additional cardiovascular risk factors, dronedarone reduced the composite endpoint of first cardiovascular hospitalization or death from any cause. The result was statistically robust and clinically meaningful in absolute terms.

The most fundamental issue is the composition of the primary endpoint. The headline 24% relative reduction in the primary endpoint is, on inspection, almost entirely a reduction in hospitalization for atrial fibrillation. The hazard ratio for atrial fibrillation hospitalization specifically was 0.63, while hazard ratios for hospitalizations driven by other mechanisms (heart failure, syncope, ventricular arrhythmia) were not significantly different. This is not surprising: dronedarone is an antiarrhythmic, it suppresses atrial fibrillation recurrence, and patients with fewer recurrences will have fewer atrial fibrillation hospitalizations. Whether this represents meaningful clinical benefit depends on how one views atrial fibrillation hospitalization as an endpoint. In an open trial in 2007, the threshold for admitting a patient with symptomatic atrial fibrillation recurrence was largely subjective. Even in a double-blind trial, an antiarrhythmic that reduces recurrence will mechanically reduce admissions for recurrence. The question is whether this prevents anything patients care about beyond the admission itself (which is often subjetive), and the trial does not answer that question.

The all-cause mortality result is what we would weight most heavily, and it was not statistically significant (HR 0.84; p=0.18). The cardiovascular mortality finding (HR 0.71; p=0.03) is a secondary endpoint in a trial with multiple comparisons, was driven by a relatively small number of arrhythmic deaths (26 vs. 48), and sits awkwardly alongside the ANDROMEDA result in which dronedarone increased mortality in patients with advanced heart failure. The authors’ framing that ATHENA excluded the high-risk heart failure patients in whom dronedarone was harmful is reasonable but not fully reassuring. Twelve percent of ATHENA patients had LVEF <45% and 21% had NYHA class II or III heart failure. The subgroup analyses did not show heterogeneity of effect, but the trial was not designed to detect harm in heart failure subgroups, and the post-trial PALLAS experience (in which dronedarone increased death and stroke in permanent atrial fibrillation) reinforces caution about how broadly to apply these results.

We are not opposed to dronedarone as a therapeutic option. It is reasonable to use in carefully selected patients with paroxysmal or persistent (not permanent) atrial fibrillation and preserved left ventricular function who are seeking rhythm control.