Author + information
- Received August 21, 2013
- Revision received February 2, 2014
- Accepted February 11, 2014
- Published online June 10, 2014.
- Leif Friberg, MD, PhD∗ ()
- ↵∗Reprint requests and correspondence:
Dr. Leif Friberg, Department of Clinical Sciences, Karolinska Institutet at Danderyd Hospital, Hjärtkliniken, Danderyds sjukhus AB SE-182 88 Stockholm, Sweden.
Objectives The aim of this study was to examine mortality and liver disease among patients exposed to dronedarone.
Background There has been concern about the safety of dronedarone, especially for patients with heart failure and permanent atrial fibrillation (AF). There have also been suspicions about liver toxicity.
Methods All 174,995 patients with a diagnosis of AF during 2010 to 2012 were identified in the Swedish Patient Register. Of these, 4,856 patients had received dronedarone according to the Swedish Drug Register, and 170,139 patients who had not were used as a control population. Mean follow-up was 1.6 years, with a minimal follow-up of 6 months.
Results Patients prescribed dronedarone were younger (age 65.5 years vs. 75.7 years, p < 0.0001) and healthier than control patients. The annual mortality rate among patients who received dronedarone was 1.3% compared with 14.0% in the control population. There were no sudden cardiac deaths and no deaths related to liver failure among patients who received treatment with dronedarone. After propensity score matching and adjustment for cofactors, patients who received dronedarone had lower mortality than other AF patients (hazard ratio [HR]: 0.41; 95% confidence interval [CI]: 0.33 to 0.51). Dronedarone patients with heart failure had lower mortality than other heart failure patients (HR: 0.40; 95% CI: 0.30 to 0.53). They also had lower mortality than expected from the general population (standardized mortality ratio: 0.67; 95% CI: 0.55 to 0.78), which indicates the selection of low-risk patients. The risk of liver disease was not increased (HR: 0.57; 95% CI: 0.34 to 0.92).
Conclusions Dronedarone, as prescribed to AF patients in Sweden, has not exposed patients to increased risks of death or liver disease.
Dronedarone is an antiarrhythmic drug for treatment of atrial fibrillation (AF). It was synthesized in search of an antiarrhythmic substance with fewer side effects than amiodarone, of which it is a derivative, and with less risk of proarrhythmia and sudden death than sotalol and Vaughan-Williams class I drugs (1–3). The safety and efficacy of dronedarone were tested against placebo in several trials in patients with nonpermanent and permanent AF, including DAFNE (Dronedarone Atrial Fibrillation Study After Electrical Cardioversion) (4), the EURIDIS/ADONIS (EURopean Trial In Atrial Fibrillation (AF) or Flutter [AFL] Patients Receiving Dronedarone for the maintenance of Sinus Rhythm and the American-Australian-African Trial With Dronedarone in Patients With Atrial Fibrillation or Atrial Flutter for the Maintenance of Sinus Rhythm) (5), ERATO trial (Efficacy and safety of dRonedArone for The cOntrol of ventricular rate during atrial fibrillation) (6), and ATHENA (A Trial With Dronedarone to Prevent Hospitalization or Death in Patients With Atrial Fibrillation) (7), as well as in patients with severe heart failure, the ANDROMEDA (European Trial of Dronedarone in Moderate to Severe Congestive Heart Failure) (8). A comparative study against amiodarone was also conducted (DIONSYSOS [Efficacy & Safety of Dronedarone Versus Amiodarone for the Maintenance of Sinus Rhythm in Patients With Atrial Fibrillation]) (9). One of the studies was stopped early because of increased mortality among patients with heart failure (8). Accordingly, the approvals by the U.S. Food and Drug Administration (FDA) and by the European Medicines Agency (EMA) in 2009 stated that dronedarone must not be used for patients with symptoms of heart failure at rest or with minimal exertion (corresponding to patients with New York Heart Association functional class IV and unstable class III heart failure).
In January 2011, after almost 150,000 patients in the United States had been prescribed dronedarone, reports about possible liver toxicity led to the issuance of a warning from the FDA. The EMA prescribed continuous surveillance with monthly liver function tests during the first 6 months. Less than a year later, the PALLAS trial (Permanent Atrial fibriLLAtion Outcome Study Using Dronedarone on Top of Standard Therapy) was stopped early because of an unexpected increase in mortality among patients with permanent AF who were treated with dronedarone (10). This prompted the FDA and EMA to issue another warning: Dronedarone should not be prescribed to patients with permanent AF.
Antiarrhythmic drugs are problematic, because they may cause dangerous ventricular arrhythmias, and side effects are frequent. Given that 2 dronedarone trials have been stopped for safety reasons, it is essential to ensure that dronedarone will have an acceptable adverse effect profile when put into daily clinical use.
The aim of this study was to determine the “real-world” safety of dronedarone, specifically with regard to all-cause mortality, mortality attributable to liver disease and heart failure, and incidence of liver disease among patients treated with dronedarone.
The first purchase of dronedarone in Sweden was made on May 24, 2010, which therefore serves as the start date of the inclusion period. The inclusion period lasted until December 31, 2012.
Patients were followed up for a minimum of 6 months with regard to all-cause mortality (until July 1, 2013). For cause-specific mortality and for incidence of liver-related morbidity, follow-up ended on December 31, 2012, because of delayed reporting from the registers.
All subjects who received dronedarone at least once in a pharmacy in Sweden during the study period constituted the study population. The control population consisted of all subjects in Sweden with a hospital diagnosis of AF who had not been prescribed dronedarone. Study patients and control patients were identified from the Swedish Patient Register and the Swedish Prescribed Drug Register. The inclusion period was May 24, 2010, through December 31, 2012. All patients given dronedarone also had a diagnosis of AF in the Patient Register.
For patients given dronedarone, the index date was defined by the first purchase of dronedarone, and time at risk was counted from that date. For control patients, the index date was defined by the first occurrence of a diagnosis of AF in the Patient Register after May 24, 2010. For the control group, we applied a blanking period of 14 days, with the result that patients who died within 14 days of the index date were excluded. Furthermore, this minimized double counting of early reappearances of diagnoses, which were often not caused by new events but were repetitions of diagnoses that caused the initial hospital stay. New diagnoses given during the first 14 days after the index date were therefore counted as being part of the baseline comorbidity, and time at risk for control patients was calculated from the index date plus 14 days. Censoring was made at the date of the event or on June 30, 2013.
All Swedish residents have a unique 10-digit civic registration number, of which the first 6 digits indicate the birth date and the ninth indicates the sex of subject. These numbers are used in all contacts with the healthcare system throughout the country, including when prescribed drugs are purchased in pharmacies. By merging information from different registers, it is possible to investigate associations between comorbidity, medication, and outcome on a national scale. To protect personal integrity, linking of registers is strictly regulated by Swedish law. When files have been combined, personal identities are replaced by anonymized numbers before the researcher obtains access to them.
The Patient Register
The Patient Register contains detailed information about hospital stays and visits to hospital-affiliated open clinics, for instance, dates of admission and discharge, principal and secondary diagnoses, and procedural and surgical codes for each hospital period. For the description of previous and current diseases, the search was limited to diagnoses given from 1997. Both principal and secondary diagnoses were used for the description of previous and current diseases and for detection of liver problems during follow-up. The mean number of diagnoses in the register increased from 1.7 in 1997 to 4.9 in 2012. The use of secondary diagnoses is less well established in conjunction with open clinic visits; the mean number of diagnoses on each contact was 0.9 in 2001 and 1.6 in 2012.
The Patient Register is validated annually by the Board of Health and Welfare in Sweden. More than 99% of all hospital discharges are technically correct (11). Generally, the validity of diagnoses in the register is good but varies among diagnoses. For example, a diagnosis of AF at hospital discharge is correct in 97% of cases (12); for most other diagnoses, the positive predictive values are in the range of 85% to 95% (13).
The Prescribed Drug Register
The Prescribed Drug Register stores details about every prescription that has been handled in every pharmacy in Sweden since July 1, 2005. Each pharmacy is required to participate by law. The register is run by a governmental agency, the Swedish Board of Health and Welfare, and is almost 100% complete because the information is transferred electronically when a drug is dispensed. It contains information about 42 variables for each purchase, including the name of the drug, its strength, quantity, dosing instruction, date of prescription, and date of purchase. We considered drugs that had been dispensed within 5 months before the index date and up to 1 month after the index date as reflecting baseline treatment. Information about drug purchases was available for the entire study period up to July 1, 2013.
Previous disease, comorbid conditions, and endpoint events were defined by coding according to the International Classification of Diseases-Tenth Revision. The specific codes used are presented in Online Table 1. Risk stratification scores for AF-related stroke (CHADS2, CHA2DS2-VASc) (14,15) and for bleeding (modified HAS-BLED) (16) were calculated for each patient by use of information from these registries (Online Appendix).
Baseline characteristics are presented descriptively, and differences were tested with Student t test and chi-square test. Incidences were calculated as events per 100 years at risk (expressed as percents in the text) and presented with Poisson rate confidence intervals (CIs). Survival is presented graphically by the Kaplan-Meier method and was analyzed with univariate and multivariate Cox regressions. In the multivariate models, we included comorbidities and medications as specified in Table 1 and Online Table 2.
We compared the observed number of deaths with the expected mortality on the basis of sex- and age-specific (1-year age groups) mortality in the general population by calculating standardized mortality ratios (SMRs). CIs were calculated with the assumption that the observed number of deaths followed a Poisson distribution.
To minimize confounding by indication (i.e., that patients selected for dronedarone treatment are different from patients not given dronedarone), propensity score analysis was used. First, a binomial logistic regression was made with factors entered that may have affected the decision to commence dronedarone treatment. From this analysis, a probability score was obtained that showed each patient's likelihood of being prescribed dronedarone. Second, patients with and without treatment for dronedarone were matched in a pairwise fashion according to their individual scores, after which multivariate Cox regression analysis was performed on a population thus made less heterogeneous.
The main analyses were performed with patients grouped according to medication at baseline, in accordance with the intention-to-treat principle. By definition, no patients in the control group could cross over to dronedarone treatment, but patients prescribed dronedarone could stop taking it during follow-up. To account for that possibility, a ratio was calculated by dividing the number of days the dispensed quantity would last by the number of days at risk. In analyses according to the on-treatment principle, patients with access to less dronedarone than needed to cover 80% of the time at risk were considered to have discontinued treatment. Values of p < 0.05 were considered significant. All analyses were performed with SPSS 22.0 (IBM SPSS Statistics, IBM Corporation, Somers, New York).
Approval for the study was obtained from the regional ethics committee in Stockholm (EPN 2010/2032-31/3, EPN 2011/690-32, and EPN 2012/1914-32).
Dronedarone was dispensed to 4,856 unique patients in Sweden between May 24, 2010, and December 31, 2012. The control group consisted of 170,139 patients who had received a diagnosis of AF during the same period but who had not received dronedarone.
Patients prescribed dronedarone were younger (65.5 years vs. 75.7 years, p < 0.0001) and healthier than AF patients not given dronedarone (Table 1), used oral anticoagulant agents more often than other AF patients (73% vs. 49%, p < 0.0001), and had tried more antiarrhythmic drugs (Online Table 2). More than one-half of the dronedarone patients had tried 1 or more class I or class III antiarrhythmic agents before they began taking dronedarone (n = 2,455 of 4,856).
Among patients who received dronedarone, the dispensed quantity was only sufficient to cover 51% of the days at risk, which indicates that the discontinuation rate was high, especially among patients who began taking dronedarone in 2010, of whom 55% had access to less dronedarone than would be needed to cover 80% of the time at risk. By 2012, that proportion had fallen to 35%.
During a median follow-up of 598 days (1.6 years), 128 patients in the dronedarone group died, which represents an annualized mortality rate of 1.3% (95% CI: 1.1% to 1.6%) compared with 14.0% (95% CI: 13.9% to 14.2%; p < 0.0001) in the control population (Fig. 1). After propensity score matching, patients given dronedarone still had a lower mortality rate than control patients (1.31% vs. 2.73%, p < 0.0001) (Fig. 2).
One-half of the dronedarone patients who died (66/128) must have stopped taking dronedarone before they died, because their supplies could not have lasted more than 25% of the time at risk. Among patients with supplies that lasted for ≥80% of the follow-up period, the annualized mortality rate was 0.43% (95% CI: 0.23% to 0.73%) compared with 1.55% (95% CI: 1.28% to 1.86%) among patients without sufficient supplies (p < 0.0001).
Information about cause-specific mortality was available until the end of 2012, during which time 95 dronedarone-exposed patients died (Online Table 3). The most common cause of death in the dronedarone group, when both principal and contributory causes were included, was cancer (38%), followed by heart failure (24%) and acute myocardial infarction (14%).
One dronedarone-exposed patient had sudden cardiac death. That patient had made a single purchase of dronedarone that was supposed to last 50 days and died 6 months later after having switched to amiodarone. There were no deaths with liver failure or liver disease listed as the principal or contributory cause of death.
Adjustment for previous and current diseases and medication did not alter the apparent survival benefit for patients taking dronedarone (HR: 0.32; 95% CI: 0.27 to 0.38) (Table 2). Propensity score matching led to some attenuation of the differences but did not eliminate the survival benefit associated with dronedarone use (HR: 0.41; 95% CI: 0.33 to 0.51).
Other antiarrhythmic drugs
Patients who used class I or III antiarrhythmic drugs at baseline had better survival than patients who did not (annual rates: 4.8% vs. 14.7%; p < 0.0001). Among users of antiarrhythmic drugs, those who used amiodarone had the highest annual mortality (8.8%) (Fig. 3). Users of dronedarone and flecainide had the lowest mortality rate (1.3% per year). After adjustment for cofactors, dronedarone use was still associated with lower mortality than amiodarone (HR: 0.40; 95% CI: 0.31 to 0.51), whereas flecainide use was not (HR: 0.67; 95% CI: 0.41 to 1.10) (Table 3). Sotalol-treated patients had the same mortality risk as amiodarone-treated patients (HR: 1.02; 95% CI: 0.77 to 1.35).
Standardized mortality ratio
AF patients who had been prescribed dronedarone had better survival than those in the general population (SMR: 0.67; 95% CI: 0.55 to 0.78), whereas patients who had not been prescribed dronedarone had approximately twice the expected mortality (SMR: 2.19; 95% CI: 2.17 to 2.21) (Table 4). For subgroups of patients with liver disease, renal failure, heart failure, vascular disease, and diabetes mellitus, the SMR was increased 3-5 fold in the non–dronedarone-exposed AF population, whereas none of these factors were significantly associated with increased mortality in the dronedarone-exposed group.
Heart failure patients
In the dronedarone group, 50 of 1,707 patients with a prior diagnosis of heart failure died during follow-up, which represents an annualized mortality rate of 2.9% (95% CI: 2.2% to 3.9%), which was barely one-tenth of the mortality among heart failure patients in the control group who did not receive dronedarone (23.9%; 95% CI: 23.6% to 24.2%; p < 0.0001). Of those who died, 39 (78%) had had access to less dronedarone than required to last for one-half of the time at risk. Dronedarone-treated heart failure patients were younger than the control subjects (66.7 ± 10.4 years vs. 79.6 ± 10.3 years; p < 0.0001) and healthier (e.g., CHA2DS2-VASc score [an AF risk score on the basis age and history of congestive heart failure, hypertension, diabetes mellitus, stroke/transient ischemic attack/thromboembolism, and vascular disease] 3.6 ± 1.6 vs. 4.9 ± 1.7; p < 0.0001). The lower death risk associated with dronedarone use among AF patients with a prior hospital diagnosis of heart failure remained significant after adjustment for cofactors (HR: 0.40; 95% CI: 0.30 to 0.53).
During follow-up, 16 dronedarone-exposed patients received a diagnosis of liver disease, which represents an annual incidence of 0.22% (95% CI: 0.12% to 0.35%) compared with 0.60% (95% CI: 0.57% to 0.64%) in the control population (p < 0.0001) (Fig. 4). After propensity score matching, the annual incidence of liver disease was no longer significantly higher than in the dronedarone group (0.22% vs. 0.37%; p = 0.104) (Fig. 5). Eleven of 16 dronedarone-exposed patients who received a diagnosis of liver disease during follow-up had not had access to enough dronedarone needed to last even half the time at risk, and 4 others had known liver problems when they started taking dronedarone. Thus, only 3 patients received a new diagnosis of liver disease while undergoing treatment, which represents an annualized rate of 0.04% (95% CI: 0.01% to 0.12%), far lower than in the nonexposed control group. The risk of receiving a diagnosis of liver disease was lower among dronedarone patients than among other AF patients (HR: 0.57; 95% CI: 0.34 to 0.92).
Patients prescribed dronedarone had a very favorable prognosis. The age- and sex-adjusted overall mortality was even lower than in the general population, in which relatively few people have AF and even fewer take dronedarone. The registers showed that many patients who were given dronedarone had previously been diagnosed with heart failure, which now is considered to be an absolute contraindication to treatment; however, the study shows that these patients fared much better than nonexposed heart failure patients. Patients taking dronedarone had fewer liver problems than patients who were not taking the drug.
High mortality in the control group
The finding of a mortality rate among dronedarone patients that was less than one-tenth that of other AF patients (1.3% vs. 14.0%) was in part caused by a high death rate in the control group, more than twice as high (SMR: 2.2) as in the general population. Most other studies have shown an approximately 1.5-fold increase in mortality for AF patients (17–20).
Patients taking dronedarone were younger and had less comorbidity than patients not taking dronedarone, but even after adjustment for no fewer than 39 cofactors found in the registers, dronedarone use was still associated with a large survival benefit. Obviously, there are other cofactors that affect prognosis that registry data do not address.
Dronedarone is intended for patients with symptomatic, nonpermanent AF. This targets dronedarone to patients with a relatively active lifestyle, which is associated with a better prognosis. The stratification of patients according to their likelihood of receiving dronedarone (propensity score matching) did compensate for some of that hidden confounding but could not eliminate it.
Relation to other antiarrhythmic drugs
Prescription data from registers are not suitable for head-to-head comparisons between antiarrhythmic drugs, because patients have not been randomized and data about drug exposure are incomplete. Interpretation of Table 3 must therefore be made with caution. Randomized trials have not shown that antiarrhythmic drugs used for AF offer any survival benefits (2,21). The finding of reduced mortality among users of antiarrhythmic drugs in the present study is a reflection of the systematic selection of low-risk patients for treatment with antiarrhythmic drugs.
This is certainly true for flecainide initiation in Sweden, where most patients are subjected to bicycle ergometer tests, echocardiography, and electrocardiographic monitoring during the first days of therapy to ensure there is no structural heart disease that could lead to an increased risk of proarrhythmia. Initiation of dronedarone is not associated with the same rigorous testing procedures. Flecainide and dronedarone share the same low mortality rate in crude numbers. This is most likely a reflection of more careful selection of low-risk patients for flecainide than for dronedarone. After adjustment for cofactors, dronedarone but not flecainide was associated with significantly lower mortality than amiodarone, despite less extensive patient-selection procedures.
In Sweden, amiodarone is used primarily as the last resort by patients for whom other drug therapy has been tried without success, that is, patients who have progressed further in their arrhythmic disease and thus have poorer prognosis. Higher mortality among amiodarone users than among users of first-line drugs is therefore natural.
Sotalol has long been, and still is, the most commonly used antiarrhythmic drug for AF patients in Sweden after beta-blockers. One would expect that a drug that is used as first-line therapy would be prescribed for patients at lower risk than a drug used after everything else has been tried. However, the annual mortality rate of 7.8% placed sotalol at the same level as amiodarone, which could indicate that sotalol use could be harmful, as indicated by a recent meta-analysis of randomized, placebo-controlled trials published by the Cochrane collaboration (2).
Heart failure patients who were given dronedarone had much lower mortality than heart failure patients who were not. This is contrary to the findings in ANDROMEDA, in which patients randomized to dronedarone had increased mortality. However, the patients in the present study who had been categorized as having heart failure were probably very unlike the patients in ANDROMEDA, in which the inclusion criteria were that there should have been at least 1 episode of shortness of breath on minimal exertion or at rest (New York Heart Association functional class III or IV) or paroxysmal nocturnal dyspnea within the month before admission (8). In keeping with the general picture that patients given dronedarone in Sweden are handpicked, low-risk patients, it is reasonable to believe that patients with heart failure who were given dronedarone had a less severe form of heart failure than heart failure patients in the control group. In ATHENA, which included patients with milder degrees of congestive heart failure, subgroup analyses demonstrated beneficial effects, in agreement with the overall study results also seen in these patients (7).
Liver disease was less common among patients exposed to dronedarone than among those not exposed to dronedarone. No liver diagnosis of any kind appeared as a cause of death or a contributing cause of death for any patient who had ever been given dronedarone. This does not rule out the possibility that some patients may have had transient elevations of liver enzymes that caused cessation of treatment without resulting in a diagnostic code, nor does it exclude the possibility that a very small number of patients may have idiosyncratic reactions that could not be detected in a study of this size. Notably, some 150,000 patients had been prescribed dronedarone in the United States before 2 cases of rapidly progressing liver failure occurred that prompted the FDA and EMA to issue their warnings about possible liver toxicity.
Exposure to dronedarone
Many patients prescribed dronedarone stopped taking it after a while. The dispensed quantities simply did not match the time at risk of these patients. When dronedarone first became available, there were many patients who had tried several antiarrhythmic agents without success and switched to dronedarone as a last resort when it became available. These patients may have failed to improve with dronedarone too and therefore stopped taking it. In support of this view, it is noted that 60% of the patients who were prescribed dronedarone in 2010 had tried 1 or more class I or III antiarrhythmic agents before they began taking dronedarone compared with 43% in 2012.
Having access to a drug is not the same as taking it. Patients who did not return for more dronedarone may have stopped taking it before they ran out of supply. An analysis of patients according to the intention-to-treat principle means that some patients will be analyzed as taking dronedarone even though they did not. This dilutes the data about any harmful effects of dronedarone and makes it more difficult to detect such harmful effects. As a safeguard against this, data were also analyzed according to the on-treatment principle, whereby only patients with access to enough dronedarone to last at least 80% of the time at risk were counted as dronedarone patients. The results of this analysis, however, were even more in favor of an advantage of dronedarone use, because those who discontinued treatment had higher mortality than those who continued treatment (Table 2). Within the group of patients who had been selected for dronedarone treatment and thus were less susceptible to confounding by indication, a higher degree of exposure to dronedarone was associated with better, not worse, outcome. A possible reason for this could be that patients who were later to die experienced treatment failure with dronedarone more often than others or were therefore taken off treatment more often.
Some of the limitations with the study methodology have already been pointed out: confounding by indication, lack of information about lifestyle factors, lack of information about the degree of disease, and uncertainty about the actual drug exposure. The positive predictive value of the Patient Register is relatively good (13), but little is known about the negative predictive value for most diagnoses, because this requires knowledge about true prevalences of diseases in the population, including subjects who have not yet received a diagnosis. The number of secondary diagnoses used is relevant to the possibility of detecting comorbidity from the register. The poorer the health of a patient and the more competing diagnoses there are, the less likely it is that codes will be given for diseases that are not the primary cause for the contact or that generally are considered as less acute and severe. A patient who has a hospital contact for chest pain and heart failure, for instance, but also has hypertension is likely to be listed without a diagnosis of hypertension if only 1 or 2 diagnoses are given.
Underdiagnosis is probably less common with diagnoses that relate to discrete events, for instance, stroke or myocardial infarction, than for diagnoses that relate to ongoing conditions such as hypertension or obesity. Underestimation of comorbidity is therefore far more likely than overestimation. Thus, confounding by indication will tend to exaggerate differences between patients taking and not taking dronedarone, whereas overestimation of drug exposure and underestimation of comorbidity will tend to reduce differences between exposed and nonexposed patients. Finally, the results of the study may not be applicable in countries where the prescription patterns for dronedarone differ substantially from the conditions in Sweden.
Dronedarone, as prescribed to AF patients in Sweden, has not exposed patients to increased risks of death or liver disease.
This research was funded by the Swedish Heart and Lung Foundation and the Stockholm County Council. Dr. Friberg has participated in advisory boards and/or given lectures for Bayer HealthCare, Boehringer Ingelheim, Bristol-Myers Squibb, Pfizer Inc., and Sanofi-Aventis; Karolinska Institute has received grants in support of Dr. Friberg's research from Boehringer Ingelheim, Sanofi-Aventis, Bristol-Myers Squibb, and Bayer HealthCare. The present report has not received financial support from the manufacturer of dronedarone, Sanofi-Aventis.
- Abbreviations and Acronyms
- atrial fibrillation
- confidence interval
- European Medicines Agency
- U.S. Food and Drug Administration
- hazard ratio
- standardized mortality ratio
- Received August 21, 2013.
- Revision received February 2, 2014.
- Accepted February 11, 2014.
- American College of Cardiology Foundation
- Touboul P.,
- Brugada J.,
- Capucci A.,
- Crijns H.J.,
- Edvardsson N.,
- Hohnloser S.H.
- Davy J.,
- Herold M.,
- Hoglund C.,
- et al.,
- for the ERATO Study Investigators
- Le Heuzey J.Y.,
- De Ferrari G.M.,
- Radzik D.,
- Santini M.,
- Zhu J.,
- Davy J.M.
- ↵The National Board of Health and Welfare. Kodningskvalitet i Patientregistret, slutenvård 2008 [Quality of Coding in the Swedish National Patient Register, In-Hospital Care 2008]. Available at: http://www.socialstyrelsen.se/register/halsodataregister/patientregistret/inenglish. Accessed April 11, 2014.
- Benjamin E.J.,
- Wolf P.A.,
- D'Agostino R.B.,
- Silbershatz H.,
- Kannel W.B.,
- Levy D.
- Andersson T.,
- Magnuson A.,
- Bryngelsson I.L.,
- et al.