OUP user menu

★ editor's choice ★

Increased mortality in paroxysmal atrial fibrillation: report from the Stockholm Cohort-Study of Atrial Fibrillation (SCAF)

Leif Friberg , Niklas Hammar , Hans Pettersson , Mårten Rosenqvist
DOI: http://dx.doi.org/10.1093/eurheartj/ehm308 2346-2353 First published online: 1 August 2007


Aims Whether paroxysmal atrial fibrillation (PxAF) affects survival is poorly recognized. Results have been conflicting in the few previously published studies. To describe mortality in patients with PxAF and to identify risk factors amenable to treatment.

Methods and results All patients (n = 2824) treated for atrial fibrillation during 2002 at one of Scandinavia's largest hospitals were followed prospectively for a mean of 4.6 years. Information about type of AF, comorbidity, and medication was acquired from medical records and national registers. Information about deaths was obtained from the National Cause of Death Register. One-third (n = 888) of the patients had PxAF (mean age 73 years). During follow-up, 267 of them died. The mean annual mortality rate was 7%. Compared with the general population, the standardized mortality ratio (SMR) was 1.6 (95% CI 1.4–1.8) for all-cause mortality, 2.4 (95% CI 1.4–3.7) for death from myocardial infarction, and 2.6 (95% CI 1.3–5.2) for death from heart failure. Warfarin treatment was associated with improved survival both in comparison with the general population (SMR 1.1 with warfarin, SMR 2.2 without warfarin) and after propensity score matching for odds to receive warfarin (HR 0.5, 95% CI 0.3–0.9). The improvement of survival could not be explained by stroke reduction alone.

Conclusion PxAF is associated with increased mortality, which mostly appears to be related to concomitant cardiovascular risks. Treatment with warfarin is associated with improved survival in PxAF patients.

  • Paroxysmal atrial fibrillation
  • Cohort study
  • Mortality
  • Risk factor
  • Anticoagulation


In contrast to permanent atrial fibrillation (AF), it is unclear whether paroxysmal atrial fibrillation (PxAF) is associated with increased mortality or not. In the few studies performed, the results have been conflicting. In a study on 143 PxAF patients in Olmsted County, Minnesota, increased mortality was seen in PxAF patients when compared with the general population,1 whereas another investigation of 418 patients with PxAF from the UK General Practice Database reported no increased risk when compared with subjects without AF after adjustment for risk factors and co-morbidity, except possibly in patients with heart failure.2 Poor prognosis in PxAF has also been reported in patients with acute myocardial infarction3 and chronic heart failure.4

Other studies, without distinguishing between type of AF and mostly concerning patients with permanent AF, have shown increased mortality when compared with the general population after adjustment for cofactors.59


The aim of this study is to describe the total and cause-specific mortality in patients with PxAF and to identify amenable contributing risk factors.

Subjects and method

All patients with AF or flutter treated as in- or outpatients at the South Hospital in Stockholm, Sweden, during 2002 were identified by ICD-10 diagnosis codes I48.9 and I49.5 with subcodes from the local patient register. The South Hospital in Stockholm serves approximately 600 000 inhabitants and is one of the largest hospitals in Scandinavia.

In addition, we identified all AF patients at Gustavsberg Primary Care Centre, which is one of the largest in Sweden, with a catchment area of 33 000 inhabitants.

The initial number of patients identified was 3405 at the South Hospital in Stockholm and 240 at Gustavsberg Primary Care Centre. The complete digitally stored information was examined and validated by two cardiologists according to a predefined protocol. Only patients with a documented episode of AF or flutter during 2002 were included. Most of the exclusions concerned duplicate identities from patients who on different occasions had been given different diagnostic codes, e.g. I48.9 on one occasion and I49.5 on another occasion, or had been treated both at the hospital and at the primary care centre on different occasions. Another frequent cause for exclusion was that patients with sick sinus syndrome without AF had been given the same ICD-code I49.5 as patients with tachybradycardia syndrome with AF. The nature of the arrhythmia behind the code had to be sorted out by reading records and studying ECG tracings. We also excluded temporary visitors to Sweden and 88 patients who died during the index-generating hospital period. For the remaining 2824 patients we obtained complementary information about previous and current disease from the National Hospital Discharge Register (HDR) going back to 1987 nationally and for Stockholm County to 1972.

The cohort was followed prospectively with annual re-examinations of medical records and through national registers of hospital discharges and mortality.

Current medication up to July 2005 was obtained from the medical records at the last contact of each calendar year. Information about the continuity of treatment was incomplete as only approximately half of the patients turned up at the hospital each subsequent year. In a previous report from the SCAF study,10 we have shown that the use of warfarin was lower than recommended, especially among patients with PxAF, and that decisions regarding warfarin treatment were affected by individual risk assessment only to a minor degree.

In the analysis, we have chosen to use data from the most recent contact for each patient as we consider this the most relevant representation of the medication at the time of death.

From the national population register, we have information on all deaths that occurred up to the end of 2006 for all patients. As for causes of death, we used the National Cause of Death Register, which at the time of analysis had been updated to 1 January 2004. Approval for the study was obtained from the Regional Ethical Committee.


The definition of PxAF used in the ACC/AHA/ESC Guidelines11 is that it is AF that terminates spontaneously within 7 days, and that termination with pharmacological or direct-current cardioversion does not change this designation. We used that definition while extracting information from the medical records, but during the process we found that exact information about the duration of the current episode of AF was often lacking. In order to avoid ambiguity in the distinction between PxAF and persistent AF, we therefore considered all patients with recurrent AF who underwent cardioversion, or where cardioversion was planned, as having persistent AF. Permanent AF was defined as longstanding AF where attempts to restore sinus rhythm had failed or else been considered as without prospects of being successful. For patients with first occurrence of AF during 2002, we used the accumulated information available by the end of that year for the classification. All patients were analysed according to the AF type in 2002 irrespective of changes during follow-up.

Echocardiography had been performed in more than two of three patients, which complemented the information from the HDR about valvular disease. For the stratification of patient risk, we used the CHADS2 scoring-system which was designed for stratification of stroke risk in AF patients.12,13 In the CHADS2 score one point is assigned for each of the risk factors: congestive heart failure, hypertension, age ≥ 75 years, diabetes, and stroke (which is given two points).

Statistical methods

We compared the observed number of deaths in patients with AF to that expected based on the mortality in the general population by calculating standardized mortality ratios (SMR). The SMRs were calculated taking gender, age (1 year age groups), and calendar year into account. SMR was calculated for total mortality as well as for specific causes using the ICD-code classifications. For the SMR we calculated confidence intervals assuming that the observed number of deaths followed a Poisson distribution.

Analyses of associations between type of AF and mortality within the cohort of AF patients were performed crude and taking risk factors and medication into account using multivariable Cox regression adjustment.

In order to control for confounding factors affecting decisions regarding warfarin treatment, we used propensity score analysis.14,15 First, we performed a binomial logistic regression entering factors that may have affected the decision to commence warfarin treatment. Examples of such factors are age and sex, type of AF, and concomitant disease. We also entered factors that, although they ought not to affect decision-making, might influence it. Examples of such factors were the clinic where the patient was treated and the day of the week when he or she was discharged. From this analysis we obtained a score for each patient showing the odds to receive warfarin. Secondly, we matched patients with and without actual warfarin treatment pair-wise according to propensity scores. Pairs of patients with and without warfarin treatment, respectively, had equal propensity scores. Finally, we performed a multivariable Cox regression analysis within this limited set of patients. All analyses were performed in SPSS 14.0.


Patient characteristics

Of the 2824 patients with AF, approximately one-third had PxAF (Table 1). The sex distribution was even. The median age (75 years) was in between that of patients with persistent (69 years) and permanent AF (80 years). The burden of risk factors and coexisting diseases was also in between that of the persistent and the permanent AF groups.

View this table:
Table 1

Patient characteristics in 2824 patients in relation to type of atrial fibrillation

Lone AF, defined as AF in patients aged < 60 years without any concomitant cardiovascular disease, was rare in all groups. In the PxAF group, fibrillation was lone in 6%.

One-third of PxAF patients with indication for warfarin treatment, and lacking contraindications, were treated with warfarin at the end of the inclusion year.


During a mean follow-up of 4.6 ± 0.3 years, 267 PxAF patients died, representing a mean annual mortality rate of 7%, to be compared with 14% in permanent AF and 3% in persistent AF. The expected annual mortality rate from age- and sex-adjusted population statistics was 5%.

Comparison with the general population

The mortality in the PxAF group was higher than expected from age- and sex-adjusted specific rates in the general population (Figure 1; Table 2). The SMR for the whole group of PxAF patients was 1.6 when compared with 2.3 in permanent AF and 0.9 in persistent AF.

Figure 1

Mortality in relation to AF type in 2824 patients during mean follow-up time of 4.6 years. Patients who died during the first hospital period are excluded. Left: Kaplan–Meier plot of unadjusted mortality in relation to type of atrial fibrillation (P < 0.0001 between all curves). Right: Multivariable Cox regression plot after adjustment for cofactors (P-value for difference between paroxysmal and persistent = 0.0008; between paroxysmal and permanent = 0.0001).

View this table:
Table 2

Standardized mortality ratios in relation to type of atrial fibrillation and concomitant risks in 2824 patients followed 4.6 years (mean)

For PxAF patients without significant comorbidity, no excess mortality was found (SMR 0.9, 95% CI 0.5–1.5). In the small group of PxAF patients with strictly defined lone AF (n = 55) there were only two deaths during the follow-up (SMR 0.6, 95% CI 0.1–3.0).

When PxAF patients were stratified according to cardiovascular risk factors using the CHADS2 scoring system, we found a strong association between risk scores and mortality (Table 2; Figure 2).

Figure 2

Mortality in relation to CHADS2 score in patients with paroxysmal AF and in all AF patients. Unadjusted Kaplan–Meier plots. Definition of CHADS2 score is given in text and in the legend to Table 2.

Causes of death

For the period in which the National Cause of Death Register was available at the time of writing (2002 and 2003), 111 deaths occurred among the PxAF patients.

SMR for death from specific causes (Table 3) showed that patients with PxAF died more often than expected from myocardial infarction (SMR 2.4), heart failure (SMR 2.6), and cardiovascular disease in general (SMR 2.1). Deaths from cerebral infarction or bleedings were not significantly more common than expected (SMR 1.5, 95% CI 0.7–2.6) among PxAF patients. There were three lethal cerebral bleedings. Two of these occurred in patients not on warfarin. Other common causes of death were not found to be more frequent than expected among PxAF patients.

View this table:
Table 3

Standardized mortality ratios for specific causes of death in relation to type of AF among 2824 patients followed from index date in 2002 until 31 December 2003

Differences in mortality between atrial fibrillation types

A comparison between the different AF types within the cohort showed that PxAF was associated with a poorer prognosis than persistent AF (Figure 1; Table 4). After adjustment for differences in age, sex, comorbidity, and medication, the hazard ratio for PxAF when compared with persistent AF was 1.5 (95% CI 1.2–2.0). Compared with permanent AF, the risk was lower; hazard ratio 0.7 (95% CI 0.6–0.9). It did not appear to matter whether the arrhythmia was flutter or fibrillation.

View this table:
Table 4

Multivariable Cox regression of factors associated with all-cause mortality among 2824 patients during a mean follow-up of 4.6 years

Warfarin treatment

Approximately, one-third of the PxAF patients were treated with warfarin (Table 1). Survival was better among patients treated with warfarin than among patients not treated with warfarin in the unadjusted crude analysis (Fig 3, left) as well as after adjustment for cofactors. The hazard ratio in favour of warfarin was 0.5 (95% CI 0.3–0.9) when compared with no warfarin.

Figure 3

Left: unadjusted incidence of death in 888 PxAF patients using the information about Warfarin use from the most recent contact during the study period. Right: multivariable Cox analysis of 187 patients with PxAF and warfarin matched with the same number of patients without warfarin, all pairs having the same odds for receiving warfarin treatment according to propensity analysis. In the multivariate analysis used for the propensity score matching the following factors were associated with warfarin treatment, previous ischaemic stroke, valvular disease, pacemaker treatment, and peripheral artery disease. Factors associated with no warfarin treatment were presence of one or several contraindications to warfarin and diabetes mellitus. After propensity score matching and adjustment for risk factors the hazard ratio for warfarin treatment was 0.5 (95% CI 0.3–0.9, P = 0.016) compared with no warfarin treatment.

In order to further control for confounding factors that might have affected the choice to treat or not to treat with warfarin, we calculated propensity scores for each patient to receive warfarin using all available background information. We then matched patients on propensity scores and performed multivariable analysis on a limited matched set of 187 warfarin-treated and 187 non-warfarin-treated patients. In these analyses, warfarin-treated patients still appeared to have better survival (hazard ratio 0.5, 95% CI 0.3–0.9, P = 0.03) (Figure 3, right).

Compared with the general population, we found no increased mortality among PxAF patients who used warfarin (SMR 1.1, 95% CI 0.8–1.4) in contrast to patients who used aspirin (SMR 1.7, 95% CI 1.4–1.9) or those without any anticoagulant treatment (SMR 2.2, 95% CI 1.6–2.8) (Table 5).

View this table:
Table 5

Standardized mortality ratios in relation to embolic prophylaxis on the latest contact and type of atrial fibrillation in 2824 patients followed 4.6 years (mean)

Warfarin use was associated with lower mortality in AF of any type.


The most important finding in this study is that patients with PxAF have an increased mortality when compared with that in the general population. Secondly, this excess mortality was restricted to cardiovascular causes of death. Thirdly, treatment with warfarin seems to improve survival in PxAF patients.

Survival among PxAF patients is better than among patients with permanent AF, but clearly lower than in patients with persistent AF. This observation was consistent in both external comparisons with the general population and in internal comparisons with patients having other types of AF.

When important co-morbidity was excluded in the PxAF group, we could no longer see any excess mortality. This does not mean that the arrhythmia in itself is without detrimental effects on health. We recognize that AF may cause embolic stroke and heart failure if adequate preventive measures are not taken. However, in the absence of previously established cardiovascular disease very few patients with PxAF actually died over a follow-up time of close to 5 years representing in total over 4000 patient-years, which indicate that either the additional risk from the arrhythmia itself is very small, or alternatively that the risk may be substantial but that the medical care given has been very efficient.

We thus believe that PxAF is an indicator of possible underlying cardiovascular disease. It is likely that efforts aimed at diagnosis and treatment of such disease is an efficient way to reduce mortality in PxAF patients.

Previously two important studies have been published on mortality in PxAF. In the Olmsted County study,1 143 patients with PxAF were followed-up for a mean 2.9 years. Mortality was found to be increased both in the unadjusted analysis and after adjustment for age, heart failure, and chronic pulmonary disease. In the other study from the UK general practice database,2 no such excess mortality was found. In this study, 418 PxAF patients had been followed for mean of 2.7 years and adjustments were made for more cofactors than in the Olmsted County study. The findings of our study, which comprises more patients and runs over a longer period of time than any of the other studies, supports the findings from the UK general practice database.

A remarkably favourable prognosis of patients with persistent AF was found both in comparison with the general population and in comparison with patients having other types of AF. In the Olmsted County study, Keating et al.1 made the same observation, which made them ask whether it is the ability to maintain sinus rhythm that makes the difference. PxAF patients alternate between sinus rhythm and AF, and may spend more time in AF than in patients with persistent AF. It has previously been shown that AF patients who remain in sinus rhythm after cardioversion have lower mortality16,17 although the reason for this is unclear. Just as heart failure may lead to AF, AF may aggravate heart failure if heart rate is poorly controlled. Death from heart failure was increased in paroxysmal and permanent AF but not in persistent AF, where there were no such deaths. Furthermore, the favourable prognosis of patients with persistent AF may partly be due to a confounding effect. It may be easier for younger and otherwise healthier patients to be accepted for cardioversion, and thus be classified as persistent AF, than for older and sicker patients who are either classified as paroxysmal or permanent depending on whether AF reverts spontaneously or not.


Warfarin treatment seems to reduce mortality in PxAF. This finding was consistent both in the comparison with the general population and in the multivariable analysis within the AF cohort. This beneficial effect cannot be explained by stroke reduction alone because death due to stroke was not common enough to accommodate for such a large reduction of fatalities. Furthermore, not even one-third of the PxAF patients were treated with warfarin.

We recognize that warfarin treatment may be withheld or be contraindicated in critically ill patients, and that this may have exaggerated the beneficial effect. However, the association between warfarin treatment and a favourable outcome remained (HR 0.6, 95% CI 0.4–0.9) after adjustment for all available information about factors affecting mortality. Furthermore, the effect remained unchanged after propensity score matching regarding factors affecting warfarin treatment (HR 0.5, 95% CI 0.3–0.9). This is consistent with the AFFIRM substudy (HR 0.5, 95% CI 0.4–0.7),16 as well as with a recent Welsh study.18

The findings in our study, as well as in the two other studies, indicate that warfarin has other life-saving effects apart from protection against ischaemic stroke in PxAF patients. Previous studies have, for example, shown that Warfarin effectively reduces mortality from ischaemic heart disease19,20 which was found at baseline among more than half of all PxAF patients in this study (Table 1).

Aspirin treatment was not as clearly associated with a reduced mortality among PxAF patients (Table 5).

Considerations about study design

A strength of this study is the large number of patients representing more than 10 000 patient-years and includes all AF patients during one full year at one of the largest hospitals in Scandinavia serving one of 15 of the Swedish population. This permits detailed analyses and statistical estimates with good precision. Furthermore, information from national registers was available which complemented the information in the medical records.

A limitation of the study is that patients were not prospectively clinically examined according to a study protocol. All information was obtained retrospectively from medical records and registers. Furthermore, cardioverted patients with PxAF may have been analysed as having persistent AF owing to difficulties to determine durations of AF episodes prior to hospital admission (see Definitions above). Patients with paroxysmal AF are a heterogeneous group of patients. Frequencies of recurrences and durations of separate episodes varied. The impact of AF burden was not addressed in the present study, but may be of importance.21

Another weakness of the study was that many patients did not visit the hospital every year, which makes the information about continuity of medication incomplete. Those who had frequent contacts, and where the information about medication is complete, probably had poorer health than those who did not. This limitation does not affect the information about mortality, which is complete from national registers.

In the comparisons with the general population, we could only adjust for age and sex but not for co-morbidity because official population statistics does not present information on individual level. To reduce this problem, we stratified the AF patients by the CHADS score12,13 which proved to be highly predictive of mortality.

Clinical implications

The development of PxAF in a patient should alert the physician to the possibility of a yet undiagnosed serious underlying cardiac disease. Treatments of such underlying disease, along with warfarin therapy, are likely to be efficient ways to improve survival in patients with PxAF.


PxAF is associated with increased mortality, which mostly appears to be related to concomitant cardiovascular risks. Treatment with warfarin is associated with improved survival in PxAF patients.


This study was supported by grants from the Stockholm County Council, AstraZeneca R&D, Mölndal, Sweden, and the Capio Research Foundation, Stockholm, Sweden.

Conflict of interest: M.R. is a national coordinator in a phase II trial for an anti-arrhythmic compound manufactured by AstraZeneca and has participated in a phase II trial and in a phase III trial investigating a thrombin inhibitor manufactured by Boehringer-Ingelheim. He has also received honoraria from AstraZeneca for giving lectures in the field of anticoagulation.


View Abstract