Figure 1. Derivation of the study population. For each patient, we identified first his or her first hospitalization for atrial fibrillation (AF) recorded in the Med-Echo database from January 1999 to March 2007. There were 3 main categories for exclusions: patients whose billings in the year before the index AF hospitalization indicated that the individual could have had prevalent AF, patients whose AF was believed to be transitory because it was induced by a reversible trigger, and patients for whom the validity of the AF diagnosis was questionable.
Figure 2. Time trends in prescription patterns before and after the publication of the Atrial Fibrillation Follow-up Investigation of Rhythm Management (AFFIRM) trial results.14 Overall trends (A) and differences in the trends between cardiovascular specialists and general practitioners (GPs) (B).The time-trend analysis of prescription patterns shows a decrease in the use of rhythm control drugs after the publication of AFFIRM results. The proportion of patients with atrial fibrillation (AF) who initiated treatment with rhythm control drugs increased by about 2% per trimester (RR, 1.02; 95% CI, 1.01-1.03) in the years before the AFFIRM trial (1999-2002), but this trend disappeared after the presentation of AFFIRM results in March 2002 (RR, 0.98; 95% CI, 0.97-0.98 for years 2002 to 2007). The change in the temporal trend was statistically significant (interaction P < .001) (A) and was similar for cardiovascular specialists and GPs (B).
Figure 3. Weighted survival of patients with atrial fibrillation (AF) on rhythm vs rate control treatment. The weighted survival curves were weighted by inverse probabilities of treatment that are equivalent to the standardization of the survival curves to the whole study population.20 The deaths in the footnote are counted in the preceding 1-year interval. The number of patients at risk in the footnote are counted at the end of each 1-year interval.
Figure 4. Effect of rhythm vs rate control therapy on mortality. Point estimates and 95% CIs can be reported at selected time points during the follow-up. The adjusted hazard ratio (HR) at a corresponding point in time quantifies the relative risks of immediate death, for rhythm vs rate control drugs, among patients who were followed until that time (ie, had not died and were not censored until that time). AF indicates atrial fibrillation.
Ionescu-Ittu R, Abrahamowicz M, Jackevicius CA, et al. Comparative effectiveness of rhythm vs rate control drug treatment effect on mortality in patients with atrial fibrillation [published online June 4, 2012]. Arch Intern Med. doi:10.1001/archinternmed.2012.2266.
eMethods 1. Potential confounders: description and measurement
eMethods 2. Propensity score models
eMethods 3. Cox model with time-dependent covariates to account for treatment switches during follow-up
eMethods 4. Bias analysis
eTable 1. ICD9/10 codes used to identify atrial fibrillation and comorbidities in administrative databases
eTable 2. Rhythm and rate control drugs included in the study
eTable 3. Results from the full Cox model (ie, adjusted for all a priori selected covariates (eMethods 1) with fixed treatment measurement (at baseline) and timedependent effects of the treatment on the death hazard.
eTable 4. Distribution of rhythm/rate control drugs at the first prescription after discharge (in decreasing order of frequency)
eTable 5. Results robustness to confounding control method
eTable 6. Results Cox regression model with time-dependent covariates (eMethods 2)
eTable 7. Clinical characteristics of full population vs those with more than 5-years of follow-up (patients who experience the long-term protective effect of rhythm control drugs)
Customize your JAMA Network experience by selecting one or more topics from the list below.
Ionescu-Ittu R, Abrahamowicz M, Jackevicius CA, et al. Comparative Effectiveness of Rhythm Control vs Rate Control Drug Treatment Effect on Mortality in Patients With Atrial Fibrillation. Arch Intern Med. 2012;172(13):997–1004. doi:10.1001/archinternmed.2012.2266
Background Controversy continues concerning the choice of rhythm control vs rate control treatment strategies for atrial fibrillation (AF). A recent clinical trial showed no difference in 5-year mortality between the 2 treatments. We aimed to determine whether the 2 strategies have similar effectiveness when applied to a general population of patients with AF with longer follow-up.
Methods We used population-based administrative databases from Quebec, Canada, from 1999 to 2007 to select patients 66 years or older hospitalized with an AF diagnosis who did not have AF-related drug prescriptions in the year before the admission but received a prescription within 7 days of discharge. Patients were followed until death or administrative censoring. Mortality was analyzed by multivariable Cox regression.
Results Among 26 130 patients followed for a mean (SD) period of 3.1 years (2.3 years), there were 13 237 deaths (49.5%). After adjusting for covariates, we found that the effect of rhythm vs rate control drugs changed over time: after a small increase in mortality for patients treated with rhythm control in the 6 months following treatment initiation (hazard ratio [HR], 1.07; 95% CI, 1.01-1.14), the mortality was similar between the 2 groups until year 4 but decreased steadily in the rhythm control group after year 5 (HR, 0.89; 95% CI, 0.81-0.96; and HR, 0.77; 95% CI, 0.62-0.95, after 5 and 8 years, respectively).
Conclusions In this population-based sample of patients with AF, we found little difference in mortality within 4 years of treatment initiation between patients with AF initiating rhythm control therapy vs those initiating rate control therapy. However, rhythm control therapy seems to be superior in the long-term.
Atrial fibrillation (AF) affects approximately 250 000 Canadians1 and 2.3 million Americans, a number that will more than double by the year 2050 owing to the aging population.2 Atrial fibrillation is associated with substantial morbidity and mortality1-9 and has a complex therapy that involves rate control agents, antiarrhythmic drugs, anticoagulant drugs, and/or ablative techniques.5,10-13 Until the publication of the landmark Atrial Fibrillation Follow-up Investigation of Rhythm Management (AFFIRM) trial in 2002,14 rhythm control (converting patients with AF to normal sinus rhythm) was generally considered superior to rate control.14Since then, clinical practice guidelines consider either strategy suitable,11,13 although some physicians still prefer the rhythm control strategy.5 Whereas it was also hypothesized that rhythm control therapy might still be superior to rate control therapy in patients with congestive heart failure (CHF), the AF-CHF trial failed to support this hypothesis (hazard ratio [HR], 0.97; 95% CI, 0.80-1.17 for all-cause mortality).15 The controversy continues, with some clinicians questioning the generalizability of the AFFIRM trial results to the general population.5,16,17 Our main objective was to compare the effectiveness of rhythm vs rate control drug treatment strategies to decrease mortality in a population-based setting with long-term follow-up.
The province of Quebec, Canada, provides universal access to health care and prescription drugs for persons 65 years or older. Databases include the hospital discharge database (Med-Echo) and the physician and prescription claims databases of the Régie de l’Assurance Maladie du Québec (RAMQ). Complete RAMQ and Med-Echo records were obtained from January 1, 1999, through March 31, 2007, for all patients 65 years or older who had a Med-Echo hospitalization record with a primary or secondary AF diagnosis (see eTable 1 for diagnostic codes). Mortality data were available until December 31, 2007. The databases were linked at the individual level using encrypted identifiers.
This study used a retrospective population-based cohort design. Patients entered the cohort at their first AF hospitalization after January 1, 1999 (index AF hospitalization), and were followed until death or end of database follow-up.
The study included all Quebec residents 66 years or older newly diagnosed as having AF during a hospitalization from January 1, 1999, to March 31, 2007. All patients (57 864) were discharged alive in the community and initiated treatment with rhythm or rate control drugs within 7 days after discharge. The time window for the first prescription was selected to capture most patients with a prescription while minimizing the potential for survival bias.18 To ensure that all patients included were newly diagnosed as having AF, we excluded 20 094 patients with medical records indicating prevalent AF (eg, rhythm control drug treatment, rate control drug treatment with an outpatient AF diagnosis, warfarin medication not associated with valvular disease, or pacemaker or implantable cardioverter-defibrillator implant in the previous year) (Figure 1). After additional study exclusions (Figure 1), the final study population included 26 130 patients.
Initial treatment was determined based on the first AF prescription dispensed within 7 days after discharge. Rhythm and rate control drugs included are listed in eTable 2. Patients prescribed simultaneously rhythm and rate control drugs were classified as rhythm control users.19 The outcome of the study was all-cause mortality. Potential confounders, selected a priori, included individual, physician-, hospital-, and community-level variables. An extended list of the potential confounders, along with measurement details, is provided in the eMethods 1.
Our main analysis used a time-to-event method. Event time was defined as the time from the first postdischarge AF prescription to death. Patients who remained alive were censored at the end of follow-up. First, the unadjusted survival under the 2 treatments was compared with the Kaplan-Meier (KM) curves. Then, to adjust the KM curves for potential confounders inverse probability weights were used to estimate weighted KM curves in which the rhythm and rate control users had the same distributions for all covariates (curves are standardized to the characteristics of the combined study population).20
Inferential analyses relied on proportional hazards Cox regression models,21 from which we report adjusted HRs and 95% CIs. The main analysis attempted to replicate an intention-to-treat analysis. Accordingly, treatment was modeled as a time-fixed binary variable (ie, patients were assigned to their initial treatment throughout the follow-up, even if they later switched to the alternative treatment). To account for possible changes in the effect of initial treatment over time, we tested and modeled potential time-dependent effects of the initial treatment with a flexible, spline-based model,22,23 while adjusting for the propensity score (ie, the probability of initiating rhythm control treatment conditional on all observed covariates) (eMethods 2).24 The proportionality of hazards assumption was rejected by the likelihood ratio test (P < .001), indicating that the relative effectiveness of the 2 initial treatments varied during the follow-up.25 Thus, we reported time-varying HRs from analyses with time-fixed treatment.
To account for potential treatment changes during the follow-up we performed additional analyses in which we used time-dependent treatment measures. Here the “current treatment” was categorized based on the most recent AF prescription and previous switches (eMethods 3). The main comparison of interest, adjusted for individual covariates, was between those treated continuously with the rhythm control drugs and those treated continuously with the rate control drugs.
Additional analyses included (1) a time trend to assess changes in AF prescription patterns after the publication of the AFFIRM trial; (2) multivariable Cox proportionality of hazards model censored at 5 years to enhance comparability with AFFIRM trial,14 (3) testing the interaction between CHF history and treatment for comparison with the AF-CHF trial,15 (4) bias analyses that assessed the robustness of our results to potential unobserved confounding (described in the eMethods 4), (5) a cohort that did not exclude prior users of rhythm or rate control drugs because the rate control drugs can be used for other indications, and (6) results from the full model with all covariates (eTable 3). The time trends in the period before and after the publication of the AFFIRM trial were estimated by Poisson models with the dependent variable defined as the proportion of patients, in a given trimester, whose initial AF treatment involved an antiarrhythmic drug. The Poisson models yield rate ratios (RR) for the linear effect of calendar trimester before and after the AFFIRM trial. A statistically significant result (P ≤ .05) when testing the interaction between the calendar trimester and the binary indicator of the period before and after the publication of the AFFIRM trial indicates that the changes in the prescription patterns that occurred after March 2002 were not by chance. All analyses were performed using SAS statistical software (version 8.02; SAS Institute Inc), with the exception of the flexible spline-based model for time-dependent effects performed using the R program.
Of the 26 130 eligible patients, 6402 (24.5%) initiated rhythm control treatment. Cardiovascular specialists were more likely than internal medicine specialists and general practitioners (GPs) to prescribe rhythm control drugs (35.9% vs 23.2% vs 21.0%). A decrease in the use of rhythm control drugs was observed after the publication of AFFIRM14 results (Figure 2A) (P < .001 for trend). A similar change in the use of rhythm control drugs occurred for cardiovascular specialists and GPs (Figure 2B).
Atrial fibrillation drugs could be prescribed either alone or in combination. The first prescription in the rate group was for β-blockers (56%), digoxin (40%), and/or calcium-channel blockers (30%). The most common initial rhythm control prescriptions included amiodarone (51%) and sotalol (24%) (eTable 4 reports all drug combinations). During follow-up, 3463 patients in the rhythm control group (54.1%) and 2210 in the rate control group (11.2%) switched at least once to the alternative treatment. The median time to switch was 108 days in the rhythm vs 124 days in the rate control group (interquartile range [IQR], 33-448 vs 48-446). The median time between the last prescription and the end of follow-up was 18 days in the rhythm control group and 19 days in the rate control group (IQR, 6-46 vs 6-54).
Distributions of cardiac and noncardiac comorbidities were similar in the 2 treatment groups. Yet, rhythm control patients were younger and more likely to have a primary AF diagnosis at admission and to undergo AF-related procedures during the index hospitalization (Table).
The mean (SD) follow-up times were 3.1 (2.3) years (maximum, 9.0 years) for all patients and 3.6 (2.5) and 3.0 (2.3) in the rhythm control and rate control groups, respectively. During follow-up, 12 632 patients (48.3%) died with 3034 (47.4%) in the rhythm control group and 9879 (50.1%) in the rate control group. The number of deaths in each 1-year interval and the number of patients at risk at the end of each 1-year interval are presented in Figure 3. For the purpose of comparison with the AFFIRM trial14 we also report crude rates for mortality (45.3%) after 5 years of follow-up (41.7% in the rhythm control group and 46.3% in the rate control group). The weighted standardized survival curves suggest no difference in mortality between the 2 treatment regimens in the first 4 years after treatment initiation but diverge later, with the rhythm control group having lower long-term risk (Figure 3).
The multivariable Cox regression, adjusted for the propensity score based on all potential confounders, showed a short-term increase in mortality for rhythm control patients in the 6 months following the treatment initiation, followed by a 3-year period with similar mortality in the 2 groups, and a steady decrease in the mortality of rhythm control patients after the fifth-year posttreatment initiation (Figure 4). The estimates at selected points in time are (1) HR, 1.07 (95% CI, 1.01-1.14) at 6 months; (2) HR, 1.03 (95% CI, 0.95-1.11) at 1 year; (3) HR, 0.95 (95% CI, 0.90-1.02) at 3 years; (4) HR, 0.89 (95% CI, 0.81-0.96) at 5 years; and (5) HR, 0.77 (95% CI, 0.62-0.95) at 8 years (Figure 4). In adjusted analyses in which we assumed a priori proportionality of hazards and censored the data at 5 years, similar to the AFFIRM14 main analyses, we found no difference in mortality between the 2 groups (HR, 1.00; 95% CI, 0.96-1.05). However, when we restricted our sample to patients who survived more than 5 years and reset the follow-up time to zero at year 5, the proportionality of hazards assumption held and the mortality was lower in the rhythm control group (HR, 0.88; 95% CI, 0.78-1.00). All results were robust to different methods of confounding control, including covariate adjustment and propensity score matching (eTable 5). Interestingly, the analysis was replicated in a study sample that did not exclude previous users of AF drugs, and, similar to the AFFIRM trial, the results did not change substantially: HR, 0.98 (95% CI, 0.95-1.01) and HR, 0.81 (95% CI, 0.75-0.87) at 1 and 8 years, respectively, after treatment initiation.
In the analyses that accounted for treatment crossovers by modeling the “current treatment,” the long-term mortality reduction was even stronger for patients who initiated and maintained rhythm control therapy relative to those who initiated and maintained rate control therapy: (1) HR, 1.01; 95% CI, 0.95-1.07; (2) HR, 0.76; 95% CI, 0.68-0.85; and HR, 0.61 (95% CI, 0.50-0.75) at 1, 5, and 8 years after treatment initiation (eTable 6). The interaction between CHF and the initial treatment tested in a model that assumed a linear time-dependent effect of treatment adjusted for covariates was not statistically significant (P = .20).
We assessed, in a large population-based study, the comparative effectiveness of initiating rhythm vs rate control drug therapy in reducing mortality in patients with AF. We found that with increasing follow-up time the mortality among the patients who newly initiated rhythm control therapy gradually decreased relative to those who initiated rate control drugs, reaching 23% reduction after 8 years of follow-up.
The recent trials comparing rhythm control vs rate control drug therapy16 concluded that there are no differences in mortality between the 2 treatment strategies.14 In contrast, our observational results in a population-based sample of hospitalized patients with AF suggest that a strategy of rhythm control is associated with lower long-term mortality. Interestingly, secondary analyses of the AFFIRM trial for cause-specific mortality and with time-dependent “on-treatment” analyses also found that rhythm control may offer a significant advantage over rate control if rhythm control can be safely achieved.26-28
There are several factors that could explain differences in results between randomized trials and observational studies,29,30 and they are crucial in interpreting our results relative to those of the AFFIRM trial.14 Generally, patients who participate in trials are not representative of the risk profiles and adherence of patients in the general population,31,32 which may lead to different treatment estimates when the treatment effect is modified by the baseline risk of the patients.30,33 Indeed, the AF clinical trials14,16 included mainly male patients with fewer comorbidities, whereas our population included more female patients with more cardiac comorbidities (eg, almost one-third of our patients had CHF compared with approximately 8% in the AFFIRM trial).14 However, even the AF-CHF trial15 population is more similar in age-sex and comorbidity distribution to the AFFIRM trial than to our population. The extent to which the treatment effect may vary as a function of these characteristics or other, unmeasured, is not known and remains to be determined. Further studies should be designed specifically to explore the hypothesis that the effect of rhythm vs rate control varies by the baseline risk of the patient.33
Differences in therapeutic regimens could also explain the discrepancy between the results of this observational study and the clinical trials. In our study, 56% of rate control patients were prescribed β-blockers vs 47% in the AFFIRM trial,14 and 51% of rhythm control patients vs 37.5%14 in AFFIRM were prescribed amiodarone. Similarly, warfarin was less frequently used by the patients in our study (54.4% vs >85%14 in the rate control group and 60.1% vs 70%14 in the rhythm control group). It is important to note that the guidelines for warfarin use changed during the study period in response to the results of the AFFIRM trial. Finally, the crossover rates in our study were higher than the ones observed in the AFFIRM trial,14 especially for those who initiated rhythm control treatment (54.1% vs 37.5%14), confirming that the selection of patients for treatment is often different in real-life than in the setting of a clinical trial. If the main reasons for switching from rhythm to rate control therapy are the inability to maintain sinus rhythm and drug toxic effects,14 some of the crossovers from rhythm to rate control drugs could be avoided by developing new antiarrhythmic drugs with fewer adverse effects28 or by using alternative methods of achieving sinus rhythm, such as ablation (which was shown in previous studies to more effective than current medications at maintaining sinus rhythm34-36).
Differences in outcomes could not explain our results because we used outcomes similar to the ones studied in the AFFIRM trial.14 While the death rate in our population is almost double than that reported in the AFFIRM trial (49.5% of patients died during 9 years of follow-up vs 16% during 5 years of follow-up),14 the difference is largely explained by the fact that our study population is almost 10 years older, on average, than the study population of the AFFIRM trial (mean age, 79 years vs 70 years). This difference is consistent with mortality statistics in the general population of Quebec, which indicate that a 10-year difference in age among seniors is associated with more than doubling the mortality rate.37 Thus, the higher crude mortality rates in our study vs the AFFIRM trial14 may merely reflect the fact that the patients in observational studies are older and sicker than those from clinical trials.32
Some differences in the length of follow-up and analysis type that explain the differences observed between our results and the AFFIRM trial14 can be shown empirically. Indeed, in sensitivity analyses in which we mimicked even closer the AFFIRM trial14 by assuming proportionality of hazards and censoring the follow-up at 5 years, we found estimates that were very similar to those of the AFFIRM trial, but a significant reduction in rhythm control patients was observed among those who survived to 5 years (eTable 5). While there is a theoretical possibility that some unknown factor(s) predispose(s) people to both increased long-term survival and higher likelihood of being prescribed (and continuing to use) rhythm control treatment, this remains a nontestable hypothesis. Indeed, sensitivity analyses in which we compared patients who survived to 5 years with the baseline population revealed no differences in the clinical characteristics of “survivors” who were originally treated with rhythm vs rate control drugs (eTable 7). This suggests that at least part of the differences in estimates between AFFIRM and our observational study is due to the fact that we had longer follow-up and that we used a more flexible analysis that accounted for statistically significant violations of the proportionality of hazards assumption. The late mortality reduction for rhythm control patients vs rate control patients could occur if the rhythm control drugs reduce long-term morbidity and mortality by potentially preventing the progression of disease toward more persistent or permanent forms of AF and the long-term risk for embolic stroke. However, this conjecture and the corresponding hypothesis-generating findings of our time-dependent analyses should be confirmed in an independent study.
Nevertheless, the major limitation of observational studies is the ability to control for confounding because randomization remains the gold standard for confounding control in any epidemiologic study. To limit the potential for confounding, we used several strategies. We restricted our sample to incident cases without reversible underlying causes of AF, we adjusted in the analyses for a large number of potential confounders, and we conducted bias analyses (eMethods) that suggest that it is very unlikely that unobserved confounding could nullify our estimates.
In addition, the generabilizability of our results could be affected by the restrictions we imposed on the study population. Specifically, to ensure that all eligible patients have complete drug coverage during the study period and that all study patients are individuals newly diagnosed as having AF who initiated rhythm or rate control drug therapy, we included only patients 66 years or older who were hospitalized with a primary or secondary diagnosis of AF. Still, within these restrictions, our study sample is an easily identifiable group of patients that is representative for the whole province of Quebec. Furthermore, this group is likely to represent most patients in the population newly diagnosed as having AF because estimates from another Canadian province show that 81.0% of patients with AF were first diagnosed during a hospitalization or in the emergency department (51.8% and 19.2%, respectively).38
Notwithstanding these limitations, this large, observational, population-based study with long-term follow-up suggests that rhythm control therapy is similar to rate control therapy in the first 4 years after treatment initiation but may be more effective than rate control therapy long-term. The risk reduction associated with rhythm control therapy was even more pronounced among the subset of patients who maintained their initial treatment over longer periods of time, suggesting that the use of rhythm control therapy may be beneficial for patients with AF in whom antiarrhythmic drugs are effective and well tolerated. The results suggest that the development of antiarrhythmic drugs with fewer adverse effects but retained or improved efficacy may result in important gains in the survival of patients with AF. For the first 4 years after treatment initiation, our results in a population-based sample are similar to the results from the recent clinical trials.14,16 In addition, we found a tendency toward a long-term protective effect for rhythm control drugs. The long-term benefits of rhythm control drugs in AF found in this study need to be assessed in future studies.
Correspondence: Louise Pilote, MD, MPH, PhD, Royal Victoria Hospital, V Building, 687 Pine Ave W, Room V2.17, Montréal, QC H3A 1A1, Canada (firstname.lastname@example.org).
Accepted for Publication: April 15, 2012.
Published Online: June 4, 2012. doi:10.1001/archinternmed.2012.2266
Author Contributions: Dr Pilote had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Ionescu-Ittu, Jackevicius, Essebag, and Pilote. Acquisition of data: Richard and Pilote. Analysis and interpretation of data: Ionescu-Ittu, Abrahamowicz, Jackevicius, Essebag, Eisenberg, Wynant, and Pilote. Drafting of the manuscript: Ionescu-Ittu and Abrahamowicz. Critical revision of the manuscript for important intellectual content: Abrahamowicz, Jackevicius, Essebag, Eisenberg, Wynant, Richard, and Pilote. Statistical analysis: Ionescu-Ittu, Abrahamowicz, Wynant, and Richard. Obtained funding: Abrahamowicz, Jackevicius, and Pilote. Administrative, technical, and material support: Essebag and Pilote. Study supervision: Essebag, Eisenberg, and Pilote.
Financial Disclosure: None reported.
Funding/Support: This research was supported by the Canadian Institutes for Health Research (CIHR), grants MOP-8127, to Dr Abrahamowicz, a principal investigator, and MOP-84304, to Drs Jackevicius and Pilote, also principal investigators. Dr Essebag is the recipient of a Clinician Scientist Award from the CIHR. Dr Ionescu-Ittu is supported by a postdoctoral fellowship from Fonds de Recherche en Santé du Québec, and Drs Eisenberg and Pilote are national researchers for the Fonds de la Recherche en Santé du Québec. Drs Abrahamowicz and Pilote hold James McGill Research Chairs at McGill University.
Role of the Sponsors: The funding agencies had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.
Previous Presentations: This study (preliminary results) was a poster presentation at the 2010 Canadian Cardiovascular Congress; October 23-27, 2010; Montréal, Quebec; and the abstract was given as an oral presentation at the 2011 North American Congress of Epidemiology; June 21-24, 2011; Montréal, Québec (abstract published in Am J Epidemiol. 2011;173[suppl 11]:636-S).
Create a personal account or sign in to: