Author + information
- Received July 25, 2018
- Accepted August 23, 2018
- Published online February 18, 2019.
- Ryan Kipp, MDa,b,
- Mariam Askari, BSc,
- Jun Fan, MSc,
- Michael E. Field, MDd,e and
- Mintu P. Turakhia, MD, MASc,f,g,∗ ()
- aDivision of Cardiology, University of Wisconsin School of Medicine and Public Health, Madison, Wisconsin
- bDivision of Cardiology, William S. Middleton Memorial Veterans Hospital, Madison, Wisconsin
- cVeterans Affairs Palo Alto Health Care System, Palo Alto, California
- dDivision of Cardiology, Medical University of South Carolina, Charleston, South Carolina
- eDivision of Cardiology, Ralph H. Johnson Veterans Affairs Medical Center, Charleston, South Carolina
- fDepartment of Medicine, Stanford University School of Medicine, Stanford, California
- gCenter for Digital Health, Stanford University School of Medicine, Stanford, California
- ↵∗Address for correspondence:
Dr. Mintu Turakhia, VA Palo Alto Health Care System, Stanford University, 3801 Miranda Avenue—111C, Palo Alto, California 94304.
Objectives In this study the authors investigated effectiveness and safety of an initial treatment strategy with class IC or class III antiarrhythmic drugs (AAD) for newly diagnosed atrial fibrillation (AF) or atrial flutter (AFL).
Background There is limited evidence to guide optimal AAD selection for rhythm control in newly diagnosed AF/AFL.
Methods Using data from TREAT-AF (The Retrospective Evaluation and Assessment of Therapies in AF), the authors performed a retrospective cohort study of patients with AF/AFL from 2004 to 2014 and class IC or class III AAD prescription within 90 days following diagnosis. Patients with prior myocardial infarction, heart failure, or end-stage renal disease were excluded. Inverse probability treatment weighted propensity scores were used to evaluate the association of AAD class on hospitalization and cardiovascular events. To evaluate residual confounding, falsification outcomes were evaluated.
Results A total of 230,762 patients developed newly diagnosed AF/AFL during the study period. Of those, 3,973 patients (1.7%) were prescribed class IC and 6,909 (3.0%) were prescribed class III AAD. Median follow-up was 4.9 years. After inverse probability treatment weighted adjustment, class IC medications were associated with lower risk of hospitalizations for AF/AFL (hazard ratio [HR]: 0.77; 95% confidence interval [CI]: 0.73 to 0.81), cardiovascular disease (HR: 0.78; 95% CI: 0.75 to 0.81), heart failure (HR: 0.70; 95% CI: 0.64 to 0.76), and lower incidence of ischemic stroke (HR: 0.74; 95% CI: 0.65 to 0.85). Similar results were found in CHADS2 (Congestive Heart Failure, Hypertension, Age ≥75 Years, Diabetes Mellitus, Prior Stroke, Transient Ischemic Attack, or Thromboembolism) 0 or 1 and CHA2DS2-VASc (Congestive Heart Failure, Hypertension, Age ≥75 Years, Diabetes Mellitus, Prior Stroke, Transient Ischemic Attack, or Thromboembolism, Vascular Disease, Age 65 to 74 Years, Sex) 0 or 1 subgroups. Falsification analyses for outcomes of urinary tract infection, pneumonia, and hip fracture were generally nonsignificant.
Conclusions Prescription of class IC AAD as initial treatment for newly diagnosed AF/AFL, compared with prescription of class III AAD, may be associated with lower risk of hospitalization and cardiovascular events.
For pharmacological maintenance of sinus rhythm, atrial fibrillation (AF) and atrial flutter (AFL) guidelines reference the presence of structural heart disease and coronary artery disease as the primary determinants of choosing a class of antiarrhythmic drugs (AAD) (1,2). For patients with minimal or no structural heart or coronary artery disease, the drug options are broad and include both Vaughan Williams class IC and class III agents, without preference for one class over the other. However, there are limited observational studies, and no randomized trials, that compare these classes of drugs with respect to safety (3–5). We therefore compared hospitalization and cardiovascular event rates between class IC and class III AAD as an initial treatment strategy in newly diagnosed AF/AFL using data from a national health care system.
We performed a retrospective cohort study of patients treated within the Veterans Affairs (VA) Health Care System with newly diagnosed AF/AFL from October 1, 2004, through September 30, 2014, using the TREAT-AF (The Retrospective Evaluation and Assessment of Therapies in AF) cohort. We used data from both administrative claims data and electronic health records of multiple VA centralized datasets, encompassing the full denominator of patients within the VA system. These datasets include the VA National Patient Care Database with demographic, outpatient, inpatient, and long-term care administrative data; the VA Laboratory Decision Support System with laboratory results from serum creatinine values; the VA Decision Support System national pharmacy extract with patient level data on inpatient and outpatient medication administration and costs; the VA Fee Basis Inpatient and Outpatient datasets; and the VA Vital Status File with validated combined vital status from the VA, Social Security Administration, and Medicare. The methods detailing the creation of this cohort have been previously published (6,7).
From this dataset, we included patients who received a prescription for a class IC AAD (defined as flecainide or propafenone) or a class III AAD (defined as dofetilide or sotalol) within 90 days of initial AF/AFL diagnosis in the study cohort. We excluded patients with any of the following: 1) a diagnosis of AF/AFL during the previous 4 years as defined by inpatient, outpatient, or fee basis International Classification of Diseases-Ninth Revision (ICD-9) codes for AF/AFL; 2) without a second, confirmatory diagnosis of AF/AFL between 30 and 365 days after the first (index) AF/AFL diagnosis; 3) not seen face-to-face in a cardiology or primary care clinic within 90 days of the first diagnosis of AF/AFL; 4) did not receive an outpatient prescription for a class IC or class III drug in the 90 days of first AF/AFL diagnosis; 5) received a prescription for a class IC and a class III antiarrhythmic medication on the same day; or 6) died within 120 days of their first episode of AF/AFL. We also excluded patients with a contraindication to either class IC or class III AAD, including patients with baseline congestive heart failure, previous myocardial infarction, and patients with an estimated glomerular filtration rate <30 ml/min/1.733 m2 or on dialysis (Figure 1).
Endpoints and definitions
The primary exposure was receipt of either a class IC AAD or class III AAD within 90 days of the index diagnosis of AF/AFL, prescribed either as an outpatient medication or prescribed for discharge following an inpatient encounter. Patients prescribed >1 class of AAD were reclassified to represent the latest prescribed medication in the 90-day period following AF/AFL diagnosis.
Outcomes were ascertained through September 30, 2015. The primary outcome was hospitalization for AF/AFL. The secondary outcomes were hospitalization for cardiovascular disease, ischemic stroke, and heart failure. We ascertained these outcomes using validated algorithms used in prior work (6–9).
Patient baseline covariates were identified and Charlson comorbidity index calculated by identification of ICD-9 codes up to 2 years before the first diagnosis of AF/AFL, using algorithms from the Agency for Healthcare Research and Quality’s Clinical Classification System (10,11). The list of ICD-9 codes used for patient identification and baseline covariate identification is available in Online Table 1. Extending the collection window beyond 2 years did not substantially increase ascertainment of comorbidities. Using previously validated algorithms for VA data, the CHADS2 (Congestive Heart Failure, Hypertension, Age ≥75 Years, Diabetes Mellitus, Prior Stroke, Transient Ischemic Attack, or Thromboembolism) and CHA2DS2-VASc (Congestive Heart Failure, Hypertension, Age ≥75 Years, Diabetes Mellitus, Prior Stroke, Transient Ischemic Attack, or Thromboembolism, Vascular Disease, Age 65 to 74 Years, Sex) scores were calculated based on the sum of the component comorbidities (12–14). Concomitant outpatient medication administration was determined using the same methods as the primary exposure. Estimated glomerular filtration rate was calculated using the Chronic Kidney Disease Epidemiology Collaboration formula (15) and the most recent outpatient serum creatinine collected from 1 year before up to 90 days after the index episode of AF/AFL and has been previously detailed (16).
We compared baseline patient characteristics using chi-squared test for categorical variables and Student t-test for continuous variables. Cox proportional hazard analysis was applied to estimate the risk of hospitalization for AF/AFL, hospitalization for cardiovascular disease, ischemic stroke, and heart failure. Cox regression models were adjusted for age, sex, race, hypertension, stroke, heart failure, myocardial infarction, diabetes, CHADS2 score, Charlson comorbidity score, digoxin, beta-blockers, calcium channel blockers, diuretic agents, antiplatelet agents, warfarin, statin, niacin/fibrates, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers and estimated glomerular filtration rate group. A parallel propensity score analysis was performed using inverse probability weighted treatment estimator. The inverse probability weights using the propensity score were calculated using multivariable logistic regression using all baseline covariates (Table 1) for the outcome of prescription of class III versus prescription of class IC antiarrhythmic medications within 90 days of AF/AFL diagnosis. Model fit was assessed using Hosmer-Lemeshow goodness-of-fit test. After inverse propensity weighting, a separate Cox proportional hazard analysis was then performed to estimate the risk of hospitalization for AF/AFL, hospitalization for cardiovascular disease, ischemic stroke, and heart failure.
We then performed a subgroup analysis on patients with a CHADS2 scores 0 or 1 and CHA2DS2-VASc scores 0 or 1 to determine whether the primary or secondary outcomes were modified in this low-risk patient cohort, which is less likely to be residually confounded by treatment indication even after accounting for covariates. Multivariate and inverse probability weighted treatment Cox regression for all outcomes was similarly performed on these subgroups.
To investigate for the presence of residual confounding when evaluating patients prescribed class IC versus class III AAD, we performed a falsification analysis by applying the full multivariable adjusted analysis cohort, the inverse weighting for the probability of treatment full cohort, and the CHADS2 scores 0 or 1 and CHA2DS2-VASc scores 0 or 1 subgroup analyses on falsification endpoints that a priori should be expected not to be associated with the choice of AAD (17). This strategy has been used previously for AF outcomes studies (18). The falsification endpoints used for this study were urinary tract infection, pneumonia, and hip fracture.
For a sensitivity analysis, we investigated the primary and secondary endpoints using a propensity score-matched cohort. Multivariate logistic regression was used to develop a propensity score using all baseline covariates (Table 1) to predict the probability of prescription of class IC versus III AAD within 90 days of AF/AFL diagnosis. Patients prescribed class IC AAD were then matched 1:1 with patients prescribed class III AAD using nearest neighbor matching. Model fit was assessed using Hosmer-Lemeshow goodness-of-fit test and C-statistic. Cox regression analysis was performed to estimate the risk of hospitalization for AF/AFL, hospitalization for cardiovascular disease, heart failure, and ischemic stroke in the propensity score-matched cohort. To evaluate for residual confounding, the falsification analysis was also repeated on the propensity score-matched cohort using the previously identified endpoints of urinary tract infection, pneumonia, and hip fracture.
SAS (version 9.2, SAS Institute, Cary, North Carolina) and STATA (version 11, STATA Corp., College Station, Texas) were used for statistical analysis. This study was approved with waiver of informed consent by the Stanford University Institutional Review Board and the VA Palo Alto Health Care System Research and Development Committee.
Of the 230,762 patients with a new diagnosis of AF/AFL during the study period, 10,882 patients (4.7%) met study criteria and were initially prescribed a class III or class IC agent within 90 days of the first AF/AFL diagnosis (age 67.3 ± 10.6 years; 2.5% women). Compared with patients prescribed class III antiarrhythmic medications, those prescribed class IC antiarrhythmic medication were slightly younger (65 ± 11 years old vs. 69 ± 10, p < 0.0001), had lower CHADS2 scores (0.8 ± 1.0 vs. 1.1 ± 1.1, p < 0.0001), lower Charlson comorbidity index (0.8 ± 1.0 vs. 1.1 ± 1.2, p < 0.0001), and had a lower prevalence of stroke or transient ischemic attack (3.2% vs. 4.7%, p = 0.0002). A CHADS2 score of 0 or 1 was present in 72% of the cohort (Table 1). Propensity distributions on both groups had similar balance and overlap (p value for goodness of fit = 0.2129; see Online Figure 1 for histogram for propensity scores).
Outcomes following prescription of class IC versus class III antiarrhythmic medications
The median follow-up was 4.9 years. In unadjusted analyses, patients prescribed class III, compared with class IC, AAD had higher crude risks of AF/AFL hospitalization (91 vs. 78 per 1,000 patient-years; p < 0.0001), cardiovascular hospitalization (194 vs. 139 per 1,000 patient-years; p < 0.0001), heart failure (26 vs. 13 per 1,000 patient-years; p < 0.0001), and ischemic stroke (10 vs. 7 per 1,000 patient-years; p < 0.0001).
After inverse weighting for the probability of treatment, prescription of class IC antiarrhythmic medications in patients with newly diagnosed AF/AFL was associated with a lower rate of hospitalization for AF/AFL (hazard ratio [HR]: 0.77; 95% confidence interval [CI]: 0.73 to 0.81; p < 0.0001). Similarly, hospitalization for cardiovascular disease (HR: 0.78; 95% CI: 0.75 to 0.81; p < 0.0001), heart failure (HR: 0.70; 95% CI: 0.64 to 0.76; p < 0.0001), and ischemic stroke (HR: 0.74; 95% CI: 0.65 to 0.85; p < 0.0001) were lower in patients prescribed class IC medications (Table 2, Figure 2).
In the subgroup of patients with CHA2DS2-VASc scores 0 or 1 (n = 5,212), after inverse weighting for the probability of treatment, prescription of class IC medications remained associated with a lower rate of hospitalization for AF/AFL (HR: 0.73; 95% CI: 0.68 to 0.78; p < 0.0001), hospitalization for cardiovascular disease (HR: 0.70; 95% CI: 0.66 to 0.75; p < 0.0001), heart failure (HR: 0.51; 95% CI: 0.42 to 0.61; p < 0.0001), and ischemic stroke (HR: 0.68; 95% CI: 0.52 to 0.91; p = 0.008). In the subgroup of patients with CHADS2 scores 0 or 1 (n = 7,882), results were similar (Table 2).
To evaluate the likelihood of residual confounding between the treatment groups, we performed a falsification analysis. The associations between the exposure groups for the studied falsification outcomes of urinary tract infection and pneumonia were nonsignificant in the multivariate adjusted cohort, in the inverse weighting for the probability of treatment entire cohort, and in the subgroups with CHADS2 scores 0 or 1 and CHA2DS2-VASc scores 0 or 1, indicating a low likelihood of persistent bias. For the falsification endpoint of hip fracture, there was a significant association in the inverse weighting for the probability of treatment analysis but not in the multivariate analysis or in the CHA2DS2-VASc scores 0 or 1 and CHADS2 scores 0 or 1 cohorts. Overall, these data suggest a reduced likelihood of residual confounding between the treatment groups (Table 3).
For a sensitivity analysis, we investigated the primary and secondary endpoints in a 1:1 propensity score-matched cohort. Using 1:1 nearest neighbor matching, 60% of the full cohort was matched (3,280 patients in each arm; Hosmer-Lemeshow goodness-of-fit test p = 0.2129; C-statistic = 0.66). The patient cohorts were generally well matched with the exception of rates of prescription of beta-blockers and calcium channel blockers, which were more often prescribed in patients receiving class IC AAD (Table 4).
In the propensity score-matched cohort, compared with patients prescribed class III AAD, patients prescribed class IC AAD continued to have a lower rate of hospitalization for AF/AFL (HR: 0.79; 95% CI: 0.73 to 0.86; p < 0.0001), cardiovascular disease (HR: 0.77; 95% CI: 0.72 to 0.82; p < 0.0001), and heart failure (HR: 0.62; 95% CI: 0.52 to 0.73; p < 0.0001). There was no significant difference in the rate of ischemic stroke between the 2 exposure groups (HR: 0.82; 95% CI: 0.64 to 1.06, p = 0.1319) (Table 5). The falsification endpoints of urinary tract infection, pneumonia, and hip fracture were not significant, indicating a low likelihood of persistent bias in the propensity score-matched cohort (Table 6).
In this large national cohort study of patients enrolled in the VA Healthcare System, prescription of flecainide or propafenone was associated with a lower adjusted risk of hospitalizations for AF/AFL, cardiovascular disease, heart failure, and incident ischemic stroke compared with dofetilide or sotalol as an initial treatment strategy in patients with newly diagnosed AF/AFL. The association was also consistent in the subgroup with CHADS2 or CHA2DS2-VASc score ≤1 and in the sensitivity analysis, and falsification outcomes generally did falsify indicating that the observed treatment effects were not likely due to residual confounding.
Antiarrhythmic medication efficacy
Previous studies have only provided limited data to support the best initial AAD choice in a population of AF/AFL without structural heart disease. A substudy of the AFFIRM (Atrial Fibrillation Follow-Up Investigation of Rhythm Management) trial identified no difference in the rate of AF recurrence or proarrhythmic events between patients on sotalol versus class I antiarrhythmic medications during a mean follow-up of 3.8 years (19). In contrast to the present study, the AFFIRM substudy focused on the efficacy of the antiarrhythmic medications at maintaining sinus rhythm rather than rate of hospitalization or ischemic stroke following medication administration. Similarly, in the PAFAC (Prevention of Atrial Fibrillation after Cardioversion) study, there was no difference in the rate of AF recurrence following cardioversion observed between those randomized to sotalol versus those randomized to quinidine and verapamil (20), and in the SOPAT (Suppression of Paroxysmal Atrial Tachyarrhythmias) trial, there was no difference in the rate of AF recurrence after 1 year of follow-up between patients randomized to either sotalol or quinidine with verapamil following a paroxysmal episode of AF (21). Both studies enrolled relatively low numbers of patients and had a short duration of follow-up, limiting ability to detect differences in outcomes.
Association of hospitalization, heart failure, and ischemic stroke with antiarrhythmic medication choice
A previously performed study using MarketScan (IBM, Armonk, New York) commercial claims data investigated hospitalization rates following AAD prescription in a low-risk patient population (age <65 years old with no history of coronary artery disease or heart failure). There was no difference in the adjusted risk of AF hospitalization, cardiovascular hospitalization, or all-cause hospitalization with sotalol compared with class IC medications (4). There are several differences in study design that may account for the disparate results. First, although the MarketScan analysis was intended to be a new user, new disease cohort, the look-back period was only 6 months (versus 2 years for the present study), which may have enriched the cohort with patients previously diagnosed with AF. The MarketScan analysis also had shorter follow-up with a median follow-up of 409 days compared with 4.9 years in the present study of an integrated health care system, allowing us to potentially capture more cumulative events. We also investigated both sotalol and dofetilide as class III antiarrhythmic medications, as opposed to sotalol alone.
The exact reasons for the reduced observed risk with class IC AAD on the clinical endpoints used in this study is unclear. Restoration of sinus rhythm in patients with previously diagnosed heart failure and AF has previously been shown to improve left ventricular ejection fraction and heart failure symptoms (22–24), suggesting an association between the occurrence of AF and worsening heart failure. In the DIAMOND (Danish Investigations of Arrhythmia and Mortality on Dofetilide) and ATHENA (A Trial With Dronedarone to Prevent Hospitalization or Death in Patients With Atrial Fibrillation) trials, use of AAD in AF was associated with reduced rate of hospitalization (25,26). Greater effectiveness at maintaining sinus rhythm with class IC compared with class III AAD may explain the difference in hospitalization and cardiovascular events observed in this study. Another possible explanation for this difference includes an increased risk of hospitalization due to class III AAD adverse events, such as bradycardia or torsade de pointes, or the need for hospitalization during reinitiation of class III AAD following inadvertent discontinuation.
Falsification analysis helps to assess vulnerability to unmeasured confounding. By analyzing outcomes that should not be associated with the exposure (class IC or class III AAD) but which could be associated with factors influencing treatment selection (frailty or chronic illness severity), we can help to isolate the effect of confounding in treatment assignment. In the present study we found no differences in risk of urinary tract infection or pneumonia between the 2 exposure arms. There was a significantly lower risk of hip fracture in the inverse weighted model but not the multivariate model. In the sensitivity analysis performed on a propensity score-matched cohort, we continued to observe a lower rate of hospitalization for AF/AFL, cardiovascular disease, and heart failure with prescription of class IC AAD, but we identified no difference in the rate of ischemic stroke between the exposure arms. The low event rate for ischemic stroke in the propensity score-matched cohort may account for this finding. Overall, these results suggest that unmeasured confounding is not likely to account for the observed treatment effect.
First, this study is an observational and retrospective cohort study from the VA Healthcare System. Unmeasured confounding cannot be completely ruled out, but it is less likely based on results of the falsification analysis and similar findings between the inverse probability of treatment weighted propensity score analysis and propensity score-matched sensitivity analysis. Most notably, we could not account for AF/AFL disease severity of burden, even though findings were similar in the subgroups with low CHADS2 and CHA2DS2-VASc scores, which are generally expected to have less severe AF/AFL. Echocardiogram data was also not available, and we therefore could not ascertain ejection fraction, although the CHADS2 and CHA2DS2-VASc 0 or 1 subgroup analyses likely reflect a population with a low prevalence of heart failure. These observational results should still be viewed as hypothesis generating. The study cohort consisted primarily of male veterans, which may limit generalizability to women or nonveteran patients. In addition, the study population had a low comorbidity burden (based on Charlson comorbidity index, CHADS2, and CHA2DS2-VASc), which may limit generalizability. Third, we used an intention-to-treat design to address the safety and effectiveness of an initial strategy selection and did not account for time-varying drug exposures or covariates. Finally, although falsification endpoints generally did falsify, particularly with in the propensity score-match cohort, the low event rates of some endpoints used in the analysis could lead to nonsignificant associations even in situations where confounding by indication was present.
In this large, retrospective cohort of patients with newly diagnosed AF/AFL, treatment with flecainide or propafenone, compared with dofetilide or sotalol, as an initial AAD selection may be associated with a lower risk of hospitalization and cardiovascular events in newly diagnosed AF/AFL. Similar findings in the inverse probability of treatment weighted propensity score analysis, low CHADS2 and CHA2DS2-VASc subgroup analysis, and in the propensity score-matched sensitivity analysis, as well as the primarily nonsignificant results in the falsification analyses reduce the likelihood that the associations are due to residual confounding between the exposure arms. Randomized trials are required to be confirmatory.
COMPETENCY IN MEDICAL KNOWLEDGE: Use of flecainide or propafenone, compared with dofetilide or sotalol, as an initial rhythm control treatment strategy for newly diagnosed AF or AFL may be associated with a lower risk of hospitalization and cardiovascular events.
TRANSLATIONAL OUTLOOK: Prospective randomized trials are needed to compare the outcomes following prescription of flecainide or propafenone versus dofetilide or sotalol for rhythm control in newly diagnosed AF or AFL.
Dr. Turakhia has received research grants from Janssen Pharmaceuticals, Medtronic Inc., AstraZeneca, Veterans Health Administration, Cardiva Medical Inc., Apple, American Heart Association, and Bristol-Myers Squibb; other research support from AliveCor Inc., Amazon, Zipline Medical Inc., iBeat Inc., iRhythm Technologies Inc.; and honoraria from Abbott, Medtronic Inc., Boehringer Ingelheim, Precision Health Economics, iBeat Inc., Akebia, Cardiva Medical Inc., and Medscape/theheart.org; and serves as an advisor to iRhythm Technologies Inc. and AliveCor Inc.
All the authors attest they are in compliance with human studies committees and animal welfare regulations of the authors’ institutions and Food and Drug Administration guidelines, including patient consent where appropriate. For more information, visit the JACC: Clinical Electrophysiology author instructions page.
- Abbreviations and Acronyms
- antiarrhythmic drug
- atrial fibrillation
- atrial flutter
- confidence interval
- hazard ratio
- International Classification of Diseases-Ninth Revision
- Veterans Affairs
- Received July 25, 2018.
- Accepted August 23, 2018.
- January C.T.,
- Wann L.S.,
- Alpert J.S.,
- et al.
- Allen LaPointe N.M.,
- Dai D.,
- Thomas L.,
- Piccini J.P.,
- Peterson E.D.,
- Al-Khatib S.M.
- Lafuente-Lafuente C.,
- Valembois L.,
- Bergmann J.F.,
- Belmin J.
- Turakhia M.P.,
- Hoang D.D.,
- Xu X.,
- et al.
- Turakhia M.P.,
- Santangeli P.,
- Winkelmayer W.C.,
- et al.
- Turakhia M.P.,
- Ziegler P.D.,
- Schmitt S.K.,
- et al.
- Hellyer J.A.,
- Azarbal F.,
- Than C.T.,
- et al.
- ↵Tarlov E. Applying Comorbidity Measures Using VA and Medicare Data. 2011 VIReC Database and Methods Cyber Seminar Series. Hines, IL: VA Information Resource Center.
- Yang F.,
- Hellyer J.A.,
- Than C.,
- et al.
- Lip G.Y.H.,
- Skjoth F.,
- Nielsen P.B.,
- Kjaeldgaard J.N.,
- Larsen T.B.
- Affirm First Antiarrhythmic Drug Substudy Investigators
- Marrouche N.F.,
- Brachmann J.,
- Andresen D.,
- et al.,
- for the CASTLE-AF Investigators
- Prabhu S.,
- Taylor A.J.,
- Costello B.T.,
- et al.
- Di Biase L.,
- Mohanty P.,
- Mohanty S.,
- et al.
- Pedersen O.D.,
- Bagger H.,
- Keller N.,
- Marchant B.,
- Kober L.,
- Torp-Pedersen C.