Author + information
- Received August 2, 2009
- Revision received December 17, 2009
- Accepted January 25, 2010
- Published online April 27, 2010.
- Mario Talajic, MD*,* (, )
- Paul Khairy, MD, PhD*,
- Sylvie Levesque, MSc*,
- Stuart J. Connolly, MD†,
- Paul Dorian, MD‡,
- Marc Dubuc, MD*,
- Peter G. Guerra, MD*,
- Stefan H. Hohnloser, MD§,
- Kerry L. Lee, PhD∥,
- Laurent Macle, MD*,
- Stanley Nattel, MD*,
- Ole D. Pedersen, MD¶,
- Lynne Warner Stevenson, MD#,
- Bernard Thibault, MD*,
- Albert L. Waldo, MD**,
- D. George Wyse, MD, PhD†† and
- Denis Roy, MD*
- ↵*Reprint requests and correspondence:
Dr. Mario Talajic, Research Center, Montreal Heart Institute, 5000 Bélanger Street, Montreal, Quebec H1T 1C8, Canada
Objectives The goal of this study was to evaluate the relationship between the presence of sinus rhythm and outcomes in patients with a history of congestive heart failure (CHF) and atrial fibrillation (AF).
Background The value of sinus rhythm maintenance in patients with AF and heart failure (HF) is uncertain.
Methods A total of 1,376 patients with AF, ejection fraction ≤35%, and heart failure symptoms were randomized to a rhythm- or rate-control strategy. Detailed efficacy analyses were used to evaluate the independent effects of treatment strategy and the presence of sinus rhythm on cardiovascular outcomes.
Results Overall, 445 (32%) patients died and 402 (29%) experienced worsening HF. The rhythm-control strategy was not predictive of cardiovascular mortality (hazard ratio [HR]: 0.90, 95% confidence interval [CI]: 0.70 to 1.16; p = 0.41), all-cause death (HR: 0.86, 95% CI: 0.69 to 1.08; p = 0.19), or worsening HF (HR: 0.86, 95% CI: 0.68 to 1.10; p = 0.23). In analyses devised to isolate the effect of underlying rhythm, sinus rhythm was not associated with cardiovascular mortality [HR: 1.22, 95% CI: 0.80 to 1.87; p = 0.35), total mortality [HR: 1.11, 95% CI: 0.78 to 1.58; p = 0.57), or worsening HF [HR: 0.62, 95% CI: 0.37 to 1.02; p = 0.059).
Conclusions A rhythm-control strategy or the presence of sinus rhythm are not associated with better outcomes in patients with AF and CHF.
Some epidemiologic studies suggest that atrial fibrillation (AF) is an independent predictor of mortality in patients with congestive heart failure (CHF) (1). Restoration and maintenance of sinus rhythm (rhythm-control) are often attempted in the hope of reducing mortality and preventing (HF) recurrences. This strategy was recently compared with a simpler “rate-control” strategy in the randomized AF-CHF (Atrial Fibrillation and Congestive Heart Failure) trial (2). Cardiovascular mortality was similar in the 2 treatment arms, whereas the rate-control strategy was associated with fewer hospitalizations and cardiac procedures.
In the AF-CHF trial, 15% of patients crossed over from one strategy to the other. Moreover, at any given time, normal sinus rhythm was present in 30% to 40% of patients assigned to the rate-control strategy, whereas AF at least once in 58% of patients randomized to rhythm-control strategy. It remains to be determined whether maintenance of sinus rhythm in patients with HF and a history of AF portends a survival advantage (as contrasted with assignment to a rhythm-control strategy). We performed a detailed efficacy analysis of the AF-CHF trial to evaluate the relationship between survival, actual treatment strategy, and prevalent cardiac rhythm in patients with AF and CHF.
In the AF-CHF trial, 1,376 patients with symptomatic HF, left ventricular ejection fraction ≤35%, and recent AF were randomized from 123 centers to either the rhythm- or rate-control strategy (2). Patients were followed at 3 weeks, 4 months, every 4 months for 48 months, and every 6 months thereafter. At each visit, a medical history and electrocardiogram were obtained, and the investigator noted whether the assigned treatment strategy had been changed. If ECG-documented AF occurred between visits, investigators noted whether it was intermittent or continuous.
Objectives of the current study were 2-fold: 1) to conduct an on-treatment efficacy analysis; and 2) to isolate the effect of underlying rhythm on clinical outcomes. Similar to the original intention-to-treat analysis (2), cardiovascular mortality was considered the primary outcome. Secondary outcomes consisted of all-cause mortality and worsening HF. The latter was defined as the occurrence of HF requiring an overnight hospital or emergency department stay or leading to a change in treatment strategy. All outcomes were adjudicated by an events committee blinded to the treatment assignment.
Continuous variables are summarized by mean ± SD. Categorical variables are represented by frequencies and percentages. Two-group baseline comparisons were performed by using the Student tor chi-square test where appropriate. Event-free survival was plotted using the Kaplan-Meier method and compared by log-rank statistics. Univariate and multivariate Cox regression models (described in later text) considered 22 static baseline and 4 time-dependent covariates, listed in Table 1.For time-dependent medical therapy covariates, a patient was considered to have used a drug if it was recorded during the evaluation preceding the interval of interest. Variables that were significant at the 0.2 level in univariate analyses were included in backward-selection multivariate Cox models and retained if associated with p values <0.01. Analyses were performed with SAS software version 9.1 (SAS Institute, Cary, North Carolina).
For on-treatment analyses, patient time was ascribed to the treatment category actually received as opposed to one that is randomly assigned (as in an intention-to-treat approach). The following 2 analyses were performed.
Efficacy Analysis by Initially Assigned Treatment
Patients were censored at the time of crossover from the randomly allocated treatment strategy. Because systematic differences may exist between patients who crossover from one direction to the other, disparities in baseline characteristics were adjusted for in multivariate Cox regression analyses. To address biases introduced by right-censoring patients who crossed over, a separate analysis restricted the study population to the 85% of patients who remained on their initially assigned treatment strategy throughout the study.
Time-Dependent On-Treatment Efficacy Analysis
Treatment strategy was modeled as a time-dependent covariate in an analysis that included all patients, without censoring at the time of crossover. With this approach, patient time was attributed to the treatment strategy in the time period under observation. Multivariate Cox regression models that included treatment strategy as a time-dependent exposure variable were adjusted for the covariates listed in Table 1(excluding the covariate AF vs. sinus rhythm).
Cardiac rhythm analysis
To isolate the effect of underlying rhythm irrespective of treatment strategy, the primary analysis consisted of modeling presence of AF versus sinus rhythm as a time-dependent covariate. Time intervals between visits were divided into quartiles. Sinus rhythm or AF was assigned to each of 4 equal time segments for every patient on the basis of ECG documentation and the investigators' determination of AF occurrence between visits, as portrayed in Figure 1A.Cox proportional hazards models, with underlying rhythm represented as a time-dependent covariate, considered the variables listed in Table 1. In addition, for each patient, the estimated proportion of time spent in sinus rhythm was calculated by dividing the total time in sinus rhythm by the follow-up duration. Patients were divided into 2 groups based on whether their proportion of time spent in sinus rhythm was equal or superior (high prevalence) or inferior (low prevalence) to the median value. Kaplan-Meier product limit curves were plotted for illustrative purposes.
All 1,376 patients (82% male, mean age 67 ± 11 years) were included in this analysis and followed for an average of 37 ± 19 months. Overall, 445 (32%) patients died, with 80% of deaths classified as cardiovascular. At least 1 episode of worsening heart failure occurred in 402 (29%) patients.
Efficacy Analysis by Initially Assigned Treatment
Over the course of the study, 208 (15%) patients crossed over at least once from their assigned strategy. Overall, 144 (21%) of 682 patients randomized to the rhythm-control strategy and 64 (9%) of 694 randomized to the rate-control strategy crossed over to the alternate treatment strategy. A single crossover occurred in 190 patients, whereas 18 patients crossed over more than once (2 to 5 times). An on-treatment analysis of cardiovascular mortality is depicted in Figure 2.In Figure 2A, the analysis is restricted to the 1,168 patients who never crossed over, whereas all 1,376 patients were included in Figure 2B, with censoring at the time of first crossover. In each case, there was no significant difference in cardiovascular mortality between groups.
Time-Dependent On-Treatment Efficacy Analysis
The results of multivariate regression analysis incorporating treatment strategy are shown in Table 2.As a time-dependent variable, the current rhythm versus rate control treatment strategy was not predictive of cardiovascular mortality, total mortality, or worsening HF.
Cardiac rhythm analysis
Analyses of underlying cardiac rhythm were limited to the 1,316 (96%) patients for whom ECGs were available. This population was followed for 4,181 patient-years, with a median time between visits of 3.7 months. In primary time-dependent efficacy analyses devised to isolate the effect of underlying rhythm on outcomes, AF was not predictive of cardiovascular mortality, total mortality, or worsening HF. The results of these analyses are displayed in Table 3.The proportion of time spent in sinus rhythm was asymmetrical, with a median of 61%, and most patients clustered either below 20% or above 80% (Fig. 1B). The clinical characteristics of patients with a high (≥61%; median 92%) versus low (<61%; median 12%) prevalence of sinus rhythm are summarized in Table 4.Figure 3depicts the absence of significant differences in cardiovascular mortality, total mortality, and worsening HF in patients with a high versus low prevalence of sinus rhythm.
The AF-CHF trial was powered and designed primarily for an intention-to-treat effectiveness analysis. As with most large randomized trials, complementary analyses may shed light on issues of efficacy. For example, treatment crossovers may dilute the beneficial effects of a therapy and occurred at least once in 15% of patients during the course of the AF-CHF trial (similar to previous rate- vs. rhythm-control trials) (3). In addition, the limited efficacy of maintaining sinus rhythm and the potential harmful effects of current therapy for AF may explain the lack of benefit of a rhythm-control strategy, even if sinus rhythm were associated with a better outcome. In the current analysis, we focused on 2 important components of efficacy: the comparative efficacy of defined rhythm- versus rate-control strategies and the impact of sinus rhythm on outcomes.
Because benefits of randomization are not fully carried over to efficacy analyses, our statistical approach combined various complementary methodologies to address potential selection biases and confounders. Censoring patients at the time of crossover or entirely excluding patients who crossed over yielded similar outcomes in patients randomized to either strategy. Moreover, in an analysis that controlled for potential confounders and made full use of the dataset (i.e., without left truncation or right-censoring), treatment strategy modeled as a time-dependent variable was not associated with mortality or worsening HF.
Cardiac rhythm analyses also failed to demonstrate an independent impact of underlying rhythm on outcomes. When the population was partitioned into 2 groups according to the proportion of time spent in sinus rhythm (i.e., high [median 92%] vs. low [median 12%]), no differences in outcomes were found. In a multivariate analysis that controlled for potential confounders and modeled underlying rhythm as a time-dependent variable, the presence of AF was not predictive of cardiovascular mortality, total mortality, or worsening HF. Clinical factors previously associated with poor survival in patients with heart failure were confirmed by our study (4), including older age, coronary artery disease, advanced New York Heart Association functional class, prolonged QRS duration, and previous cerebrovascular events. In addition, we found that oral anticoagulation therapy was associated with improved survival, consonant with proven protective effects in patients with AF and risk factors for stroke (5).
Although our results seem to contrast with findings from AFFIRM (Atrial Fibrillation Follow-up Investigation of Rhythm Management) (6), inconsistencies may reflect, in part, the different study populations, treatment strategies, and means of ascertaining time-dependent exposures. In AFFIRM, sinus rhythm modeled as a time-dependent covariate was associated with a better prognosis and antiarrhythmic drug use with a poorer outcome. Importantly, the AF-CHF trial was restricted to patients with severe systolic dysfunction and symptomatic HF, whereas few patients in AFFIRM had left ventricular dysfunction. In addition, almost all patients in the rhythm-control arm of the AF-CHF trial received amiodarone, whereas other antiarrhythmic drugs were used in AFFIRM, including sodium channel blockers, which have previously been associated with increased mortality (7). Finally, our study design permitted a more comprehensive classification scheme for time spent in AF, incorporating 2 sources of information. In contrast, AFFIRM relied exclusively on ECGs obtained at each visit. Previous studies of HF have provided inconsistent results, some suggesting that AF is associated with a poorer prognosis (1). These discrepancies may be due, in part, to differences in study populations with variable incidences of AF and distributions of prognostically important factors such as age, left ventricular ejection fraction, and comorbidities that may confound the relationship between AF and outcomes. The most plausible explanation for our findings is that AF is a marker of more advanced disease but that it is not etiologically linked to poorer outcome. Indeed, patients with a higher AF burden had a longer history of AF, more persistent forms, larger left atrial dimensions, and worse New York Heart Association functional class. This suggests a greater degree of structural changes associated with heart disease, perhaps explaining why AF per se was not independently associated with outcome. These findings are consistent with observations by Yamada et al. (8) who demonstrated slower atrial conduction and elevated atrial natriuretic factor in HF patients in whom AF subsequently developed. Amiodarone was used to maintain sinus rhythm in the vast majority of our patients, thus reflecting practice in the many constituencies across 4 continents in which our study was conducted. Recent reports of pulmonary vein ablation in selected patients with HF suggested important improvements in ventricular function (9,10). However, these series have been small and lacked hard clinical outcomes. Larger randomized trials with appropriate end points are needed to fully assess the invasive strategy in patients with HF.
Efficacy analyses have inherent limitations that arise from loss of the randomization effect. We used a multifaceted analysis with complementary approaches devised to minimize systematic bias. Importantly, the consistency of our analyses suggests that our findings were robust to specific approaches. It is well-known that episodes of AF may be silent, leading to underestimation of the true AF burden (11). However, this potential limitation does not apply to on-treatment analyses, underscoring the importance of the various approaches to assessing efficacy.
A rhythm-control strategy is not associated with better outcomes in patients with AF and CHF, even when analysis is performed to consider efficacy in sinus rhythm maintenance. Clinical implications are substantial, consistent with all previous clinical trials comparing rate- versus rhythm-control strategies and support the notion that rate control is an acceptable primary treatment option in a sizable proportion of the population with AF with symptomatic HF and irreversible left ventricular systolic dysfunction.
Canadian Institutes of Health Research.
Dr. Talajic received consulting fees from Medtronic, Boehringer Ingelheim, and Sanofi-Aventis, and grant support from Boehringer Ingelheim, Bristol-Myers Squibb, Servier, Medtronic, and Sanofi-Aventis. Dr. Khairy received consulting fee and lecture fees from St. Jude Medical. Dr. Hohnloser received consulting fees from Sanofi-Aventis, Cardiome, and Aryx, and lecture fees from Sanofi-Aventis. Dr. Macle received consulting fees from Biosense-Webster, Sanofi, and St. Jude Medical. Dr. Stevenson received consulting fees and research support from Medtronic. Dr. Waldo received consulting or lecture fees from Biosense-Webster, Sanofi-Aventis, AstraZeneca, Bristol-Myers Squibb, CardioInsight, Biotronik, St. Jude Medical, Daiichi Sankyo, Ablation Frontiers, and GlaxoSmithKline, and grant support from Bristol-Myers Squibb, CV Therapeutics, and Sequel Pharmaceuticals. Dr. Wyse received consulting fees or paid advisory board membership from Boehringer Ingelheim, Medtronic, Sanofi-Aventis, Bristol-Myers Squibb, Biotronik, Boston Scientific/Guident, and Biosense-Webster. Dr. Roy received consulting or lecture fees from Astellas, Cardiome, and Sanofi-Aventis.
- Abbreviations and Acronyms
- atrial fibrillation
- congestive heart failure
- confidence interval
- heart failure
- hazard ratio
- Received August 2, 2009.
- Revision received December 17, 2009.
- Accepted January 25, 2010.
- American College of Cardiology Foundation
- The AFFIRM Investigators
- Flaker G.C.,
- Blackshear J.L.,
- McBride R.,
- et al.
- Yamada T.,
- Fukunami M.,
- Shimonagata T.,
- et al.
- Page R.L.,
- Wilkinson W.E.,
- Clair W.K.,
- et al.