OUP user menu

Atrial fibrillation is associated with increased mortality: causation or association?

Darryl P. Leong, John W. Eikelboom, Jeff S. Healey, Stuart J. Connolly
DOI: http://dx.doi.org/10.1093/eurheartj/eht044 1027-1030 First published online: 13 February 2013

This editorial refers to ‘All-cause mortality in 272 186 patients hospitalized with incident atrial fibrillation 1995–2008: a Swedish nationwide long-term case–control study’, by T. Andersson et al., on page 1061

It is important to determine whether the excess mortality observed in patients with atrial fibrillation (AF) is directly due to AF or is just an association. Not only do patients want to know if AF is a cause of premature death, but knowing that there is or is not a causal relationship will influence therapeutic choices. Such knowledge will also certainly impact on the AF research agenda for the future. If AF directly causes excess mortality, then the use of therapies that specifically and successfully eliminate AF—rather than just prevent its symptoms—are preferable. Andersson et al. have now presented their findings from a retrospective observational study of patients with incident AF compared with AF-free controls identified through national databases.1 Similar to previous studies (Table 1), they report that AF is an independent risk factor for all-cause mortality. Although the study by Andersson et al. was very carefully conducted and is one of the largest ever performed in patients with AF, can we conclude from these observational data that AF is a direct cause of premature death?

View this table:
Table 1

Studies examining the relationship between atrial fibrillation and all-cause mortality

StudyCohortAdjusted relative risk of death (95% CI)
Andersson1272 186 incident AF<65 years:65–74 years:75–85 years:
vs. 544 344 controlsF 2.15F 1.72F 1.44
M 1.76M 1.36M 1.24
Ruigómez61035 chronic AF vs.F 2.8 (2.2–3.6)
5000 controlsM 2.3 (1.8–3.0)
Benjamin7621 incident AF vs.F 1.5 (1.2–1.8)a
1242 controlsM 1.1 (0.9–1.4)a
Conen81011 incident AFF 1.70 (1.30–2.22)
vs. 33 711 controls
Haywood9334 prevalent AF2.01 (1.68–2.41)
vs. 30,370 controls
Stewart10100 prevalent AFF 2.2 (1.5–3.2)
vs. 15 306 controlsM 1.5 (1.2–2.2)
Miyasaka114618 incident AF2.08 (2.01–2.16)
vs. general
  • AF, atrial fibrillation; CI, confidence interval; F, female; M, male

  • aOdds ratio of death by pooled logistic regression amongst those surviving at least 30 days after AF diagnosis.

To address this question critically, we will use the conceptual framework proposed by Sir Austin Bradford Hill2 in which the lines of evidence for causality are: strength of association, consistency of association, biological gradient (a ‘dose–response relationship’), biological plausibility, and experimental evidence (Table 2).

View this table:
Table 2

Modified Hill criteria2 for establishing causality

CriterionEvidence to support AF causing excess mortality
Strength of associationModerate
Consistency of associationStrong
Biological gradientWeak
Biological plausibilityModerate
Experimental evidenceAgainst
  • AF, atrial fibrillation.

Strength and consistency of association

Andersson et al. reported an adjusted relative risk for all-cause mortality amongst patients with incident AF of ∼1.5–2, with higher relative risks in younger individuals and females.1 This represents a moderately strong association.3 To put this relative risk into context, in a 40-year prospective cohort study, smoking exhibited a similar hazard ratio of ∼1.5 for all-cause mortality.4 In a meta-analysis of the association between cigarette smoking and lung cancer, a relative risk of 5.5 was reported.5 Hence the association between AF and mortality is consistent with a causal one, but not so strong as to constitute strong evidence of causation.

Studies evaluating the association between AF and death have been consistent in reporting an independent relationship between AF and mortality (Table 1);611 with relative risks ranging from ∼1.5 to 2.5, although the elevated risk is not present in all subgroups in some studies. Despite adjustment for co-morbidities, all the studies are observational and subject to the potential for confounding. Further, there may be bias favouring the publication of positive associations, at the expense of neutral studies which could inflate the consistency of the association seen between AF and mortality.

Biological gradient

There is some evidence to suggest a biological gradient for the association between AF burden and all-cause mortality. The Women's Health Study showed no significant association between incident paroxysmal AF and death, whereas any type of AF (including persistent and permanent) was independently associated with an increased risk of mortality.8 In contrast, data from the Loire Valley AF Project demonstrated no increase in mortality risk in individuals with persistent or permanent AF, compared with paroxysmal AF.12

Biological plausibility

The biologically plausible mechanisms by which AF might cause death include thrombo-embolic events and worsening of heart failure (induced by tachycardia, or possibly by beat-to-beat ventricular irregularity). However, stroke appears to account for a very small proportion of the deaths in AF patients. In Olmsted County, the most common causes of death in patients with incident AF were coronary artery disease (15%), heart failure (16%), malignancy (14%), and stroke (7%).11 Thus it appears that stroke deaths are not the most common modes of death in AF patients. While AF may have caused a proportion of heart failure, it is just as likely that AF resulted from worsening heart failure. In the AF-CHF study, reducing AF in heart failure patients did not prevent worsening heart failure or heart failure deaths.13 Although stroke mortality explains a part of the increase in death seen with AF, we do not have a clear biological mechanism that explains most of the observed increase in mortality.

Experimental evidence

According to Bradford Hill, it is through an experiment that ‘the strongest support for the causation hypothesis may be revealed’.2 The definitive human experiment to test for causality is the randomized clinical trial, and we have several well designed and executed clinical trials of AF suppression which do not support the hypothesis that AF itself increases mortality. The AFFIRM study was a randomized mortality trial examining the effect of a strategy aimed at restoring and maintaining sinus rhythm in 4060 AF patients.14 This trial showed no survival benefit in the rhythm control group over the rate control group. As already mentioned, the AF-CHF study showed no mortality benefit from a rhythm control strategy in 1376 patients with both heart failure and AF.13 One potential reason for the lack of benefit to rhythm control strategies is the deleterious effects of antiarrhythmic drug therapy, which can have both cardiac and extra-cardiac toxicities. Dronedarone, which has a similar electropharmacological profile to amiodarone, has been shown to lower both AF burden15 and cardiovascular mortality in patients with non-permanent AF.16 Despite these initial promising results, the PALLAS study demonstrated that dronedarone increased mortality in a higher risk cohort of patients with permanent AF.17

Catheter ablation is the most effective strategy to reduce AF, yet at present we do not know if it will reduce mortality because we lack adequately powered studies. Randomized trials of AF ablation that are powered for mortality may help to clarify whether the association is causal, but will require careful interpretation as benefits of AF suppression with ablation may be offset in part by the complications of the procedure.

Could factors associated with atrial fibrillation explain increased mortality?

If it were true that AF does not directly lead to increased mortality, how do we explain the numerous well conducted observational studies which have reported independent increases in mortality with AF?

The answer may be that these studies, despite using sophisticated statistical methods for adjustment of baseline differences, cannot adjust for unmeasured confounders. Examples of potential confounders that were not evaluated in these observational studies include myocardial (left ventricular) fibrosis, concomitant digoxin use and toxicity, obesity, obstructive sleep apnoea, control of hypertension, and patient adherence to heart failure and other therapies. Some or all of these factors can cause AF and may also increase risk of death (Figure 1).

Figure 1

The relationship between atrial fibrillation and mortality. Atrial fibrillation may directly lead to death, or there may be confounding factors that predispose to both atrial fibrillation and death. OSA, obstructive sleep apnea.

Conclusions and future directions

Consistent evidence indicates that AF is associated with increased mortality, but the extent to which this is a direct effect of AF itself or is related to the numerous serious associated conditions remains a puzzle. It is likely that AF itself directly increases the risk of death in some patients; and it is also a marker of worsening of heart failure, hypertension, valvular disease, and other associated conditions. Improving our understanding of how AF is associated with mortality may offer the potential for new treatments that reduce the risk of premature death. Both AF itself and some of its associated conditions are modifiable, and it is important to know where to direct scarce healthcare resources. Thus, we believe that future research should focus on understanding the mechanisms by which AF is associated with increased mortality and developing strategies to modify these mechanisms. Present-day clinical efforts should emphasize the prevention of complications of AF that are associated with excess mortality, such as thrombo-embolism and uncontrolled ventricular rate, and the treatment of AF co-morbdities that also contribute to premature mortality. AF should be recognized as a marker for increased mortality and should prompt clinicians to target affected patients aggressively with proven secondary prevention strategies related to hypertension, heart failure, and obesity.

Conflict of interest: none declared.


  • The opinions expressed in this article are not necessarily those of the Editors of the European Heart Journal or of the European Society of Cardiology.


View Abstract