Article Text


Digoxin in atrial fibrillation: report from the Stockholm Cohort study of Atrial Fibrillation (SCAF)
  1. L Friberg1,
  2. N Hammar2,3,
  3. M Rosenqvist4
  1. 1Department of Clinical Science and Education, Karolinska Institute at South Hospital, Stockholm, Sweden
  2. 2Institute of Environmental Medicine, Karolinska Institute, Stockholm, Sweden
  3. 3AstraZeneca R&D, Mölndal, Sweden
  4. 4Department of Cardiology, Department of Clinical Science and Education, South Hospital, Karolinska Institute, Stockholm, Sweden
  1. Correspondence to Dr L Friberg, GlobenHeart Clinic, Arenavägen 33, Box 100 01, S-121 26 Stockholm, Sweden; leif.friberg{at}


Background Recent studies of patients with heart failure and of patients receiving intensive care indicate that digoxin may increase mortality if the patient has atrial fibrillation (AF).

Objective To study which patients receive digoxin treatment for AF and what the prognosis is for patients given this treatment.

Method 2824 patients with AF were studied prospectively for a mean of 4.6 years. Information about medication was obtained from the local hospital registry. Information about diagnoses, hospitalisations and deaths was obtained from national registries. Propensity score matching and Cox regression was used to account for confounding.

Results Factors associated with digoxin use were permanent AF (hazard ratio (HR) = 3.2, confidence interval (CI) 2.7 to 3.9), absence of pacemaker (HR = 2.3, CI 1.6 to 3.2), history of heart failure (HR = 2.0, CI 1.7 to 2.5), treatment in an internal medicine ward rather than a cardiology ward (HR = 1.6, CI 1.3 to 2.0), female sex (HR = 1.6, CI 1.3 to 1.9) and age ≥80 years (HR = 1.4, CI 1.1 to 1.7). More patients with than without digoxin died (51% vs 31%, p<0.001). After adjustment for covariates, however, no disadvantages related to digoxin use could be found for all-cause mortality, myocardial infarction, ischaemic stroke, time to readmission to hospital or days at hospital/year at risk. The only end point significantly associated with digoxin use was pacemaker implantations, which were more common in digoxin-treated patients (HR = 2.0, CI 1.2 to 3.4).

Conclusion Digoxin is mainly given to an elderly and frailer subset of patients with AF and is thus associated with an increased mortality. When differences in patient characteristics are accounted for digoxin use seems to have a neutral effect on mortality and major cardiovascular events in patients with AF.

View Full Text

Statistics from

Digoxin and related cardiac glycosides have been used for more than 200 years for treatment of oedema and heart failure,1 and for almost 100 years for rate regulation in atrial fibrillation (AF).2 While the drug has beneficial effects in heart failure3 and can reduce resting heart rate in AF,4 reports have indicated that use of digoxin may be an independent risk factor for death.

These studies have mostly been made on groups of patients with heart failure or other acute conditions making it difficult to extrapolate the results to general patients with AF. Thus, in a study on patients with sinus rhythm and heart failure the Digitalis Investigation Group (DIG) randomised over 6000 patients to digoxin or placebo and found that there was no difference in overall mortality, although there was a trend towards fewer deaths due to worsening of heart failure among digoxin-treated patients.5 This favourable trend was, however, counteracted by a higher incidence of arrhythmic deaths in the digoxin group.

In the AFFIRM trial6 a post hoc analysis showed an association between digitalis medication and mortality (hazard ratio (HR) = 1.42, 95% confidence interval (CI) 1.09 to 1.86).

Recently, a study of the “Registry of Information and Knowledge about Swedish Heart Intensive care Admissions” (RIKS-HIA)7 showed that a subgroup of intensive care patients with AF but without heart failure had higher mortality if they were treated with digoxin (HR = 1.42, CI 1.29 to 1.56). No such difference was seen in patients with AF and heart failure, or in patients with heart failure without AF, although mortality was considerably higher in these groups.

The study aimed at identifying which patients with AF receive digoxin treatment, and the long-term prognosis for these patients given this treatment.


The Stockholm Cohort study on Atrial Fibrillation (SCAF)8–10 comprised all 2912 patients who had received a diagnosis of AF or atrial flutter while treated as inpatients or outpatients during 2002 at the South Hospital in Stockholm, Sweden, or at the Gustavsberg Primary Care Centre in a community of 32 000 inhabitants close to Stockholm.

The patients were retrospectively identified from local registers at the hospital and at the primary care centre early in 2003. The complete digitally stored information was examined and validated according to a predefined protocol. Complementary information about previous diseases and hospitalisations was obtained from the Swedish National Hospital Discharge Register (HDR) going back to 1987 nationally and to 1972 for Stockholm County.

The cohort, thus constituted, has since then been followed up with annual reanalyses of the medical records in order to identify changes in medication, rhythm status and events. Information about incident hospitalisations and deaths during the follow-up was obtained through national registers. At the time of writing information about deaths was available up to 1 January 2007, whereas information about hospitalisations was available up to 1 January 2006. Therefore, observation time, and time at risk, varies depending on the end point in question. The mean follow-up time is 4.6±0.3 years, median follow-up was also 4.6 years.

For this study we excluded 88 patients who died during the index-generating hospital period, leaving 2824 patients for study.

Patients were analysed according to their digoxin status on the latest documented contact of the inclusion year 2002. Thus, digoxin exposure may have changed during follow-up but the patients are still analysed according to their original status. Patients who were cardioverted or had a catheter ablation procedure for AF or atrial flutter during follow-up were not excluded from analyses, but these circumstances were used as covariates in the multivariable analyses.


For the definition of end points other than death we used the respective ICD-10 codes given at the time of hospitalisation and available for all hospitals in the country through the National HDR. For identification of de novo implantations of pacemaker systems we used the procedure codes in the same register. To avoid ambiguity in the classification of persistent AF and paroxysmal AF, we considered all patients who underwent cardioversion as having persistent AF. As contraindications against warfarin treatment we considered presumed poor compliance (dementia, psychiatric illness, alcohol or drug abuse), frequent falls, previous severe bleeds, malignant hypertension, severe haemostatic defects, severe liver disease, cancer with increased risk of bleeds, recent trauma or surgical procedures. Age >90 years was considered a contraindication unless the patient had daily help with medication.

Previous and present diseases were identified from registers by their appropriate ICD codes given at the time of hospital discharge. Knowing that chronic conditions, in contrast to acute events, may escape reporting as a supplementary diagnosis at discharge, we used complementary information from the local medical records—for example, patients were categorised as diabetic even if the diagnostic code had not been used at discharge, if the medical records showed that they used antidiabetic drugs or had more than two diagnostically elevated glucose tests. Likewise, patients with more than two recorded blood pressure readings exceeding 160/95 mm Hg were categorised as having hypertension, unless the readings had been obtained while the patient was in severe distress. Echocardiography had been performed in more than two-thirds of the patients, which complemented the information from the HDR about valvular disease. Insufficiencies grades 2–4/4 were considered as valvular defects. Stenosing defects of at least moderate severity were also considered as valvular defects.

Statistical methods

For pairwise comparisons, t tests and χ2 tests were used. For survival analyses we employed Kaplan–Meier analysis with log-rank test. For multivariable analyses of survival, we used Cox proportional hazards regression having ascertained that the proportional hazards assumption was not violated by making sure that hazard ratios (HRs) were reasonably constant across time and that regression curves did not cross.

In order to take into account differences in patient characteristics associated with the probability of receiving digoxin treatment in the multivariable analyses we used propensity scores. The use of propensity scores is widely considered as a tool to increase comparability and to reduce the potential influence by confounding by indication in non-randomised studies of drug use.11 The propensity score analyses included 1342 patients, half of whom were treated with digoxin at baseline, and half of whom were not. This means that almost 85% (671/802 patients) of the digoxin-treated patients were included and thus only 15% had to be excluded owing to lack of suitable matching non-digoxin-treated subjects.

For the identification of factors associated with digoxin treatment at baseline we used logistic regression with all of the factors in table 1 in the model together with information about the clinic of discharge and the day of the week on which the patient was last discharged from hospital during 2002, or whether the patient was treated only as an outpatient. From the logistic regression we obtained scores for the likelihood of digoxin treatment for each person based on the available information about the patients current and previous comorbidities, treatment and demographic factors that was entered into the analysis. Patients with and without digoxin treatment were matched according to identical propensity scores at the level of three decimals. After comparison of mortality in a population matched for propensity score, we stratified patients into three groups of equal sizes according to whether the propensity scores were low, intermediate or high and used Cox regression, entering presumed confounders successively in more complex analyses in order to observe the effect of each confounder on the final estimate.

Table 1

Baseline characteristics of study population

All tests were two-sided. Confidence intervals (CI) were 95%. p Values <0.05 were considered as significant. All analyses were performed in SPSS version 16.0 and 17.0.


Characteristics of patients given digoxin treatment

Digoxin was prescribed to 28% of the 2824 patients with AF during the inclusion year 2002 and to 35% of the patients according to the medication status at the latest documented contact during the study period. Patients who received digoxin at index differed in important aspects from patients who did not receive digoxin (table 1). In short, a typical recipient of digoxin was an elderly woman with permanent AF, with heart failure and generally of poorer health than those who did not take digoxin.

In a multivariable analysis of factors associated with prescription of digoxin (table 2), permanent AF was the strongest determinant, followed by absence of pacemaker (ie, patients with pacemakers do not get digoxin, they get something else). Patients discharged from internal medicine wards rather than from cardiology wards were, after adjustment for cofactors, more likely to get digoxin (HR = 1.63, CI 1.31 to 2.02). They were also less likely to be cardioverted during follow-up (HR = 0.27, CI 0.17 to 0.41) or to be discharged with a statin (HR = 0.42, CI 0.32 to 0.55) or an ACE inhibitor (HR = 0.53, CI 0.43 to 0.66) than patients at cardiology wards.

Table 2

Factors associated with treatment with digoxin at baseline among 2824 patients with atrial fibrillation (AF)

Mortality and digoxin treatment

During the observation period which lasted 4.6±0.3 years 1038 patients died, of whom 412 were prescribed digoxin at index and 626 were not. The mortality was higher among patients who were treated with digoxin (51% vs 31%, p<0.001, univariate HR = 1.94 (CI 1.71 to 2.20); fig 1). When adjustment was made for age, sex, comorbidities and medication the difference was substantially reduced (HR = 1.10, CI 0.94 to 1.28). We also studied the relationship between mortality and digoxin treatment at the latest, rather than the first, contact during the observation period. Unadjusted mortality was higher among patients treated with digoxin (48% vs 31%, p<0.001) but after multivariable adjustment there was essentially no difference (HR = 1.05, CI 0.92 to 1.20).

Figure 1

Unadjusted all-cause mortality among 2824 patients with and without digoxin at base line.

When patients were matched according to their individual propensity scores there was no longer a difference in mortality related to digoxin use (univariate HR = 1.05 (CI 0.90 to 1.23); fig 2). We further divided the matched patients into three groups according to whether their chances for digoxin treatment was low, intermediate or high (table 1) and found that mortality was similar in all three groups (table 3). Finally, multivariable Cox regression was used to further compensate for differences in cofactors that had not been neutralised by propensity score matching. Still no excess mortality was found among patients with AF treated with digoxin compared with those without (HR = 1.04, CI 0.89 to 1.21).

Figure 2

Unadjusted all-cause mortality among 1342 patients matched pairwise according to the propensity score for treatment with digoxin.

Table 3

Outcome in relation to digoxin use at index

Other outcomes

Patients treated with digoxin, who had high propensity scores for this treatment, were less often hospitalised specifically for heart failure (table 3), although the number of days at hospital for any cause did not differ between groups. Nor were there any differences in the frequency of myocardial infarctions or ischaemic stroke related to digoxin use. There was, however, an increased rate of pacemaker implantations among patients with digoxin as compared with those without digoxin (HR = 1.99, CI 1.16 to 3.43).


The main finding of this study is that digoxin treatment in patients with AF appears to have a neutral effect on long-term mortality and major cardiovascular events after accounting for age, comorbidity and other patient characteristics. Thus, crude differences in mortality appear to be due to patient selection rather than to a deleterious effect of digoxin itself.

Digoxin is predominantly used by elderly patients with a higher risk, thus it is to be expected that theirs is a less favourable prognosis. Women were prescribed digoxin more often than men, regardless of age and other cofactors that we could account for. This is noteworthy since digoxin is no longer the recommended first-line treatment for management of heart rate in AF except in patients with congestive heart failure, hypotension or, possibly, in “patients who are so sedentary as to obviate the need for rate control during activity” according to the recommendations in the ACC, AHA, ESC Guidelines.12 We could not adjust for differences in lifestyle factors. Possibly, women with AF are more sedentary than men, but it may also be that doctors are more prone to assume that a man is in greater need of adequate rate control during activity in order to perform well, than a woman, especially if she is elderly.

Digoxin was less often used for patients with pacemakers or for those treated in cardiology wards as opposed to patients treated in wards belonging to the clinic of internal medicine. In a previous report8 we found that a similar pattern applied to the administration of warfarin for patients with AF between cardiologists and internists, with cardiologists acting more actively and more in accordance with current guideline recommendations than the internists.

In this study there is nothing to indicate that patients receiving digoxin had worse survival rates and more major cardiovascular events than those not receiving this treatment. The use of pacemaker implantations was about twice as high among patients with digoxin. This difference remained after adjustments for differences in cofactors. Digoxin is a relatively poor agent for rate control, without appreciable effect during exercise. It is therefore conceivable that digoxin was regarded as sufficient for rate regulation in a number of patients with undiagnosed atrioventricular conduction abnormalities who sooner or later would have been in need of a pacemaker even if they had been treated with a β blocker or an antiarrhythmic agent.

The results in this study are not easily compared with the findings in the other major studies on digoxin treatment in patients with AF since patient selection and study design differ considerably. The results or our study resemble the results from the prospective randomised DIG trial5 concerning patients with heart failure without AF. Both studies failed to detect any differences in mortality. In the DIG study there were fewer hospitalisations due to heart failure among patients treated with digoxin at baseline than among patients without digoxin. In our study this was true for patients with a high probability score for digoxin treatment, but not for patients with low or intermediate scores. For hospitalisations regardless of cause, there was no benefit from digoxin.

The post hoc AFFIRM study6 and the RIKS-HIA study7 both showed increased mortality among digoxin-treated patients. In the AFFIRM substudy, the authors interpretation was not that digoxin has a deleterious effect on survival, but that digoxin may be prescribed for patients at greater risk of death, such as those with heart failure. In the RIKS-HIA study the increase in mortality only applied to a subgroup of patients with AF, but without heart failure. Although this study included a large number of patients, it was performed in an intensive care setting which makes it difficult to translate the results. In our study, in which slightly more than half of the patients had a history of heart failure, no such association was seen.

Methodological considerations

The most important limitation to this study is that our information about the patients’ exposure to digoxin, and other medications, is fragmentary. The information about medication is complete for all patients at baseline only. During follow-up, information about medication could only be obtained when, and if, patients happened to be in contact with South Hospital. This implies that the information about patients with poor health and frequent hospital contacts is more complete than for younger and healthier patients. Medication may have changed during follow-up without being recorded in this study. The paucity of data did not permit digoxin exposure to be used as a time-dependent covariate. Instead, we analysed patients according to the treatment given as they entered the observation period and according to the medication on the latest recorded contact, respectively. Patients who are given digoxin are often given this with the intention that the treatment should go on indefinitely. Thus, of all 671 patients matched for propensity score who used digoxin in 2002, each one of them continued to receive digoxin on the latest recorded contact. Digoxin seems to be the last resort or the final choice of medication when AF is permanent and patients are too frail for anything else.

The propensity score analyses included almost 85% of the digoxin-treated patients and thus few patients were excluded owing to lack of suitable matching non-digoxin treated patients. The results from these analyses should therefore be basically generalisable to digoxin-treated patients with AF in populations resembling the Swedish SCAF cohort which has previously been well characterised.8–10

A major limitation of this study is the lack of randomisation to digoxin treatment as in the DIG study of patients with heart failure from 1997.5 We performed propensity score analysis to compensate for this and in order to increase comparability of patient characteristics between digoxin treated and untreated patients. Notably, today it would be difficult to carry out a randomised trial with digoxin use for rate control in AF, since there are more adequate means to achieve this nowadays, and since the use of digoxin as a mainstay for rate control in AF is no longer regarded as in line with current recommendations.

A strength of this study is that it is based on a cohort of fair size where the information about concomitant and previous disease and medication at baseline has been combined from several sources—that is, written medical records, protocols from the local physiology laboratory as well as from the National HDR which contains every diagnostic code given for each patients retrospectively for several decades, and prospectively for recording events occurring during follow-up. The Swedish National Population register has essentially no loss of deaths and provides a more or less complete follow-up of mortality. The HDR also has a high degree of completeness. The validity of a hospital discharge diagnosis of myocardial infarction has been evaluated repeatedly in Sweden and found to be high for epidemiological purposes.13 Similar findings have been reported for stroke, although from fewer studies and with less consistency.14 15 The validity of a hospital discharge diagnosis of congestive heart failure in Sweden has been shown to be reliable with fewer than three of the hospitalisations missed by the national register.16 17 It is likely that any misclassification of disease in this study was unrelated to digoxin treatment and therefore had little effect on the main results.


Digoxin is mainly given to an elderly and frailer subset of patients with AF and is thus associated with an increased mortality When these and other differences in patient characteristics are accounted for, digoxin use appears to be neutral for the long-term mortality and major cardiovascular events in patients with AF.


View Abstract


  • Funding This study was supported by unrestricted grants from the Stockholm County Council and AstraZeneca R&D, Mölndal, Sweden.

  • Competing interests LF has no conflicts of interests to declare. NH is a part-time employee of AstraZeneca R&D, Mölndal, Sweden. MR has been lecturing and acted as a national coordinator in clinical trials sponsored by Astra Zeneca. He is also a member the Boehringer Ingelheim Swedish advisory committee regarding haemostasis.

  • Ethics approval Approved by the regional ethics committee of Stockholm.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.