Author + information
- Received May 18, 2010
- Revision received July 26, 2010
- Accepted August 10, 2010
- Published online January 25, 2011.
- Satoru Kodama, MD, PhD⁎,
- Kazumi Saito, MD, PhD⁎,
- Shiro Tanaka, PhD†,
- Chika Horikawa, RD⁎,
- Aki Saito, RD⁎,
- Yoriko Heianza, RD⁎,
- Yui Anasako, RD⁎,
- Yukako Nishigaki, RD⁎,
- Yoko Yachi, MS⁎,
- Kaoruko Tada Iida, MD, PhD⁎,
- Yasuo Ohashi, PhD‡,
- Nobuhiro Yamada, MD, PhD⁎ and
- Hirohito Sone, MD, PhD⁎,⁎ ()
- ↵⁎Reprint requests and correspondence:
Dr. Hirohito Sone, Professor, Department of Internal Medicine, University of Tsukuba Institute of Clinical Medicine, 3-2-7 Miya-machi, Mito, Ibaraki 310-0015, Japan
Objectives The purpose of this meta-analysis is to summarize the estimated risk of atrial fibrillation (AF) related to alcohol consumption.
Background Results from observational studies examining the relationship between alcohol consumption and AF are inconsistent.
Methods A systematic electronic search of Medline (January 1966 to December 2009) and Embase (January 1974 to December 2009) databases was conducted for studies using key words related to alcohol and AF. Studies were included if data on effect measures for AF associated with habitual alcohol intake were reported or could be calculated. The effect measures for AF for the highest versus lowest alcohol intake in individual studies were pooled with a variance-based method. Linear and spline regression analyses were conducted to quantify the relationship between alcohol intake and AF risk.
Results Fourteen eligible studies were included in this meta-analysis. The pooled estimate of AF for the highest versus the lowest alcohol intake was 1.51 (95% confidence interval: 1.31 to 1.74). A linear regression model showed that the pooled estimate for an increment of 10 g per day alcohol intake was 1.08 (95% confidence interval: 1.05 to 1.10; R2 = 0.43, p < 0.001). A spline regression model also indicated that the AF risk increased with increasing levels of alcohol consumption.
Conclusions Results of this meta-analysis suggest that not consuming alcohol is most favorable in terms of AF risk reduction.
Atrial fibrillation (AF) is the most common sustained arrhythmia, representing a growing epidemic, and is accompanied by serious complications. Atrial fibrillation accounts for 45% of all embolic strokes and has a deleterious impact on longevity, with an approximate doubling of all-cause mortality (1). Although the etiology of AF is not fully understood, many epidemiological associations with AF, including both cardiac (e.g., valvular disease, cardiomyopathy, coronary artery disease) (2) and noncardiac conditions (e.g., aging, obesity, sleep apnea, diabetes mellitus, metabolic syndrome, heavy alcohol consumption) (3) have been vigorously investigated.
The association of episodic heavy alcohol use with the onset of AF has been recognized as “holiday heart syndrome” for a long time (4). Recently, it has been hypothesized that not only episodic but also habitual heavy alcohol consumption is associated with the risk of AF (5). However, results from epidemiological studies that aim to confirm this hypothesis have been inconsistent, although high alcohol consumption has been associated with several major disease groups such as neoplasms and cardiovascular diseases (6). It is also important to clarify the overall impact of any degree of alcohol intake on AF risk given that moderate alcohol consumption has been associated with a lower risk of cardiovascular disease (7) or all-cause mortality (8). Therefore, our aim of this meta-analysis of observational studies is to review the risk of AF in relation to alcohol consumption, focusing on determining if there is a dose-response relationship between any degree of alcohol consumption and the risk of AF as well as AF risk in relation to heavy alcohol drinking.
Electronic literature searches (Medline, January 1966 to December 2009; and Embase, January 1974 to December 2009) to identify studies describing alcohol intake and AF were conducted using medical subject headings related to alcohol (alcohol drinking OR alcohol related disorders OR alcoholism OR alcoholic beverage OR ethanol) and AF (arrhythmias OR atrial fibrillation). Reference lists from the identified articles were manually examined for relevant new articles. This process was repeated until no additional articles could be identified. No language restriction was imposed.
For inclusion, a study had to fulfill the following criteria: 1) have a cohort or case-control design; 2) identify AF as an outcome variable of interest separate from other arrhythmias; and 3) provide or allow calculation of the effect measure (i.e., relative risk [RR] in a prospective study or odds ratio [OR] in a retrospective study) with its corresponding confidence interval (CI). As an exception, because of the overlap between AF and atrial flutter, studies in which AF and atrial flutter were combined as a study outcome were also included. However, when data on risk of both AF only and the combination of AF and atrial flutter were provided simultaneously in 1 study, we used data on the risk of AF only. We excluded studies wherein alcohol consumption was classified as “yes” or “no” because the degree of daily alcohol consumption could not be ascertained through such a response.
Two of our investigators (S.K. and H.S.) independently reviewed all relevant articles and identified eligible studies. Discrepancies were resolved by group discussion. We extracted the following data from each publication: first author's name, year of publication, geographic region, design of the observational study (i.e., cohort or case-control), selection of study population (i.e., hospital-based or population-based), participants' characteristics (i.e., age [mean or range], proportion of men, and whether participants with heart disease that influenced AF risk were excluded), characteristics of outcome (i.e., onset or recurrence, dominantly paroxysmal atrial fibrillation [PAF] or dominantly chronic [persistent] AF, and whether atrial flutter was included in the study outcome), methods of assessment of alcohol consumption (i.e., questionnaire, interview, or reviews of medical records and registries), methods for ascertainment of AF (i.e., electrocardiogram screening, registries, or participant's report), category of alcohol intake, number of participants and cases, and study-specific controlled variables.
The effect measure in each study was extracted or, if the effect measure for AF was not provided, it was calculated based on data on the number of cases and noncases in referent and exposed groups. In principle, we defined the lowest alcohol intake category or no drinking as the referent group and the other category as the exposed group. When a study classified >2 alcohol intake categories, we extracted or calculated all available effect measures for AF. If a study provided several effect measures, such as unadjusted and adjusted effect measures, the most completely adjusted effect measure was used.
The effect measures were transformed to their natural logarithm (log OR/RR). Fundamentally, the standard error (SE) was calculated from the corresponding CI. In some studies (9–11), the SE corresponding to the log OR/RR was not provided. Then we directly calculated the SE corresponding to the log OR/RR using data on the number of cases and noncases in the exposed and referent groups in each comparison as follows: (in case of log OR) (9,11) or: (in case of log RR) (10), where C1 and N1 indicate the number of cases and noncases in the exposed group, respectively, and C0 and N0 indicate the number of cases and noncases in the referent group, respectively. If necessary, the effect measure and its corresponding SE were approximated from figures in the manuscripts using an image scanner (CanoScan LiDE 500F [resolution 600 dpi], Canon, Inc., Tokyo, Japan).
To standardize alcohol intake, we used a common scale (grams per day) for ethanol consumption. When a study used the number of drinks per day as a unit of alcohol intake, the unit was transformed into grams of ethanol according to the study-specific methods for estimating the amount of ethanol per drink. If the amount of ethanol per drink was not specified, the unit was considered equivalent to 12 g ethanol (12).
For each study, data on the mean level of daily alcohol intake for each category were extracted or calculated as point estimates of ethanol consumption. When this information was not provided, we assigned the mid-point of the upper and lower boundaries in each category as the average intake. If the highest category had an open upper boundary, mean alcohol intake was estimated to be 1.2 times the lower boundary (13).
To summarize the association of habitual heavy alcohol consumption with the risk of AF, the effect measures were pooled for the highest versus lowest alcohol intake category. Based on the definition of heavy alcohol drinking by the National Institute on Alcohol Abuse and Alcoholism (12), we limited this analysis to studies in which the highest alcohol intake category was defined as consumption of 2 or more drinks per day for men, 1 or more drinks per day for women, and 1.5 or more drinks per day for the combination of men and women. Subjects described as “alcohol abusers” and “alcoholics” were also considered to be heavy drinkers. The pooled estimate was calculated by averaging the log OR/RRs weighted by the inverse of variance based on a fixed- or random-effects model. We used the results from the random-effects model if between-study heterogeneity, which was assessed by Q statistics and I-squared (14), was significant (15). Because daily alcohol consumption in heavy alcohol drinkers varied from study to study, we conducted stratified analysis according to the degree of “heavy” drinking. Analyses were also stratified by the pre-specified study characteristics. We also conducted meta-regression analyses to assess the influence of study characteristics on study results.
The possibility of publication bias was assessed primarily by visual inspection of a funnel plot in which the effect measure in the individual study was plotted against its corresponding SE. The funnel plot is expected to be symmetrical with respect to the overall estimate if publication bias is absent. We secondarily assessed the possibility of publication bias by 2 formal tests: the Begg's adjusted rank correlation test (16) and the Egger's regression asymmetry test (17). If publication bias was statistically suspected, we also followed the Duval and Tweedie “trim and fill” procedure (18) for further estimation of the possible effect of the publication bias. This method considers the possibility of hypothetical unpublished studies that would have allowed a funnel plot to be symmetrical and recalculates a pooled estimate after imputation of the effect measures of the hypothetical studies as though they actually existed.
We primarily used weighted, least-squared regression models (19) to explore the dose-response relationship between alcohol intake and the risk of AF by regressing the log OR/RR of AF on the alcohol dose. To further investigate the shape of the relationship between the level of alcohol consumption and the risk of AF, we used restricted cubic splines with knots at the 25th, 50th, and 75th centiles of the distribution of alcohol consumption. These analyses were limited to data from studies with a referent category whose mean alcohol consumption was reported or estimated to be less than 1 drink per day, so that overlapping of alcohol intake of exposed and referent groups could be avoided as much as possible. Two-sided p values of ≤ 0.05 were considered statistically significant except for tests of publication bias for which the recommended level is p value ≤0.10 (20). Data were analyzed using STATA software version 10 (STATA Corp., College Station, Texas).
Figure 1 shows details of the literature search. Our electronic literature search resulted in retrieval of 1,754 citations (591 from Medline and 1,163 from Embase). Of these, 1,673 citations were excluded after the first screening. Eighty-one papers as well as 20 additional papers identified by manual search were left for full-text review. After this review, of the 101 papers, 87 were excluded for the reasons shown in Figure 1. Finally, 14 studies (9–11,21–31), which comprised 130,820 participants and 7,558 cases, were included in this meta-analysis.
Characteristics of the 14 included studies are shown in Table 1. Nine studies used a cohort design, 4 studies used a case-control design, and 1 study reported data from both case-control and cohort designs. All studies were conducted in Western countries (7 in Europe and 7 in North America).
For assessing daily alcohol consumption, only 4 of the 14 studies (25,26,29,31) validated methods to assess alcohol intake. All effect measures were controlled for age and sex, and most of the included reports (10 studies) made adjustments for heart disease, which potentially elevates the risk of AF, or excluded participants with heart disease. Only 3 studies (10,29,31) considered racial differences among the participants.
Risk of AF through heavy alcohol consumption
The lowest amount of alcohol consumed in the highest category in each study ranged from 1.5 to 6 drinks per day. Consequently, in all 14 studies, the criteria for heavy alcohol drinking as previously defined were met (12). Three studies reported separate results according to sex. One study analyzed 2 populations separately according to age (≤60 years or >60 years), and 1 study indicated 2 risk measures (1 for PAF and another for the progression from an AF episode to permanent AF). Finally, 19 effect measures were analyzed to summarize the risk of AF in relation to heavy alcohol intake. The pooled estimate of OR/RR for the highest category of alcohol consumption compared with the lowest category in individual studies was 1.51 (95% CI: 1.31 to 1.74) (Fig. 2).
There was significant between-study heterogeneity in the effect measures (Q-squared, 33.2; I-squared, 45.8%; p = 0.02). Table 2 shows results of stratified and meta-regression analyses across a number of key study characteristics to explore causes of the study heterogeneity. On the whole, a positive association between AF risk and heavy alcohol consumption was consistently found in all stratified analyses.
In the stratified analysis by mean alcohol intake in the highest intake group, the pooled estimates of AF for <4 drinks per day, or 48 g per day, and ≥4 drinks per day was 1.32 (95% CI: 1.15 to 1.50) and 1.74 (95% CI: 1.35 to 2.24), respectively. However, the difference was not significant (p = 0.17). When analyses were limited to the 6 studies that regarded nondrinkers as the referent group, the pooled estimate for the highest category was 1.36 (95% CI: 1.18 to 1.57).
Stratified analysis by geographic region, participants' sex, or whether persons with heart disease were included did not show any significant difference in pooled estimates between strata. Using a case-control design seemed to produce a substantially larger AF risk (pooled estimates 1.98 [95% CI: 1.49 to 1.63]) compared with the use of other designs (pooled estimate 1.34 [95% CI: 1.22 to 1.47]). However, these differences were not borderline significant (p = 0.06). Selecting hospital-based participants produced a borderline significantly larger pooled estimate in comparison with population-based participants (pooled estimate 1.75 [95% CI: 1.45 to 2.11] vs. 1.30 [95% CI: 1.18 to 1.44]; p = 0.049).
Strong associations were observed when the type of AF end point in the study was PAF-dominant (pooled estimate 1.92 [95% CI: 1.44 to 2.56]) or AF recurrence (pooled estimate 2.37 [95% CI: 1.44 to 3.90]) whereas a significantly weaker association was observed in studies that included atrial flutter as a study end point (pooled estimate 1.25 [95% CI: 1.10 to 1.43]) than in those that did not (pooled estimate 1.83 [95% CI: 1.45 to 2.30]; p = 0.02).
The method for ascertainment of AF did not significantly affect the magnitude of the association between high alcohol consumption and AF risk. However, a significantly stronger association was observed when reviewed data on alcohol intake were based on medical records or registries (pooled estimate 2.17 [95% CI: 1.74 to 2.70]) compared with other methods of determining alcohol consumption, such as questionnaires or interviews (pooled estimate 1.28 [95% CI: 1.16 to 1.41]; p < 0.001). The influence of study adjustments for possible confounders was not significant, although AF effect measures were attenuated with adjustment for hypertension or blood pressure (pooled estimate 1.33 [95% CI: 1.17 to 1.50]).
Publication bias was visually suggested by the asymmetrical funnel plot of the reported results (Fig. 3), which was also statistically supported by Egger's test (p = 0.03) but not Begg's test (p = 0.31). We attempted to adjust for this publication bias using the trim and fill method (18). After 4 negative unpublished results were incorporated to produce a hypothetically symmetrical funnel plot, the pooled estimate of AF for heavy alcohol consumption was modestly attenuated to be 1.39 (95% CI: 1.19 to 1.62) but remained statistically significant (p < 0.001).
Dose-response relationship between alcohol intake and AF risk
Nine studies (10,11,23–27,29,30) involving 126,051 participants and 6,341 cases were eligible for analysis of the dose-response relationship between different categories of alcohol intake and AF risk. The alcohol dose in these studies ranged from 4.0 to 86.4 g per day. Figure 4 illustrates the linear and spline regression curves for AF risk related to daily alcohol intake. The linear dose-response curve showed a significant relationship between alcohol intake and AF risk (R2 = 0.43, p < 0.001). The coefficient for the linear term was 7.4 ± 1.3 × 10−3, meaning that the incremental increase in relative risk of AF per 10 g alcohol consumption per day was = 1.08 (95% CI: 1.05 to 1.10). Also, in the spline regression model, AF risk significantly increased with larger daily alcohol consumption levels (R2 = 0.44, p < 0.001). However, the fit of this model was not significantly different from that of the linear regression model (p = 0.77).
Our study is the first to systematically review the literature on the association between alcohol consumption and the risk of AF. In this review, high alcohol intake was shown to be associated with a significant elevation in AF risk, both by overall analysis and across a number of stratified analyses based on key characteristics of study methods, although there was substantial study heterogeneity in the magnitude of AF risk, partly due to variability in study design and methodology. Even though, in principle, observational studies do not allow for proof of causality, there are several theoretically plausible speculations for the cause-effect relationship between high alcohol intake and the development of AF.
One speculation is based on biological findings that suggest a harmful effect of high alcohol intake on maintenance of normal heart rhythm, including the achievement of a hyperadrenergic state (32), impairment of vagal tone (33), direct effect on myocardial structure (34), and various electrophysiological changes in atrial cells (e.g., increase in intra-atrial conduction time represented by a length of the P-wave, reduction in the refractory period, negative inotropic effect through calcium-channel inhibition in ventricular cells) (31,35,36).
Another speculation is based on reports suggesting that the development of chronic heart failure accompanied by long-term excessive alcohol consumption may result in elevated AF risk (2). Particularly, dilated cardiomyopathy is typical of alcohol abusers with chronic heart failure. The average total lifetime alcohol consumption was reported to be significantly greater in patients with dilated cardiomyopathy than in a population-based control group (37). Moreover, alcoholics were found to have progressive dilated cardiomyopathy in proportion to the duration of heavy drinking even before the clinical appearance of chronic heart failure (38).
It remains to be established whether the dose-response relationship between daily alcohol consumption and AF risk is interpreted as linear or not (e.g., J-shaped curve, threshold curve). While we identified a linear association of daily alcohol consumption with the risk of AF, a J-shaped relation or threshold value was not observed from the current spline regression curve. These regression analyses suggest no evidence that moderate alcohol consumption is beneficial in ameliorating the risk of AF, unlike that of cardiovascular disease (7). If anything, moderate alcohol drinkers may have a greater risk of AF than nondrinkers, although the AF risk is not as large as that for heavy drinkers.
First, the definition of heavy drinking is heterogeneous across studies. Second, the majority of the included studies did not state whether the method used for assessment of alcohol intake was validated. Third, few studies considered racial differences among participants. Differences in ethnicity or proportion of whites and nonwhites among different studies might have affected AF risk estimates. Fourth, asymptomatic PAF could have been missed in any of these studies. Given that heavy alcohol drinkers are likely to have experienced PAF, the risk of AF in relation to high alcohol intake would be underestimated in any of the examined studies. Fifth, no studies have investigated the effect of different types of alcoholic beverages on AF risk, although it has been reported that wine has a better effect on cardiovascular disease (39). Sixth, it could not be ruled out whether a particular drinking pattern, such as whether alcohol was consumed with a specific food or at meals, could have contributed to the AF risk irrespective of alcohol dosage. For example, alcohol is usually consumed during meals, as in Mediterranean countries (39), which might explain a beneficial effect.
A meta-analysis cannot completely solve problems with confounders that vary from study to study. Lack of adjustment for possible confounders could also produce a superficially strong association between high alcohol consumption and AF risk. For example, a weaker association between high alcohol consumption and AF risk was observed when studies included blood pressure values or the presence of hypertension among study confounders. In fact, hypertension was reported to be an independent risk factor for AF (2), and the risk of hypertension increases linearly with alcohol consumption (40). Possibly, the AF risk associated with high alcohol consumption is partly explained by an alcohol-related development of hypertension. In addition, other factors that could not be specified by this meta-analysis or were not specified in the individual included studies might contribute to residual confounding (e.g., objective sleep apnea , diabetes mellitus ).
Lastly, results that indicated risk of AF could be biased by study design and other methodological features. For example, larger AF risk estimates were observed in studies having a case-control design compared with studies having a cohort design or in hospital-based studies compared with population-based studies. The AF risk could have been overestimated by exaggeration of alcohol intake in patients with AF in studies with a case-control design (i.e., recall bias) or an unavoidable reduction in alcohol intake as a result of a control subject having an illness in a hospital-based setting (i.e., selection bias). However, underestimation of AF risk is possible if patients did not truthfully report the full extent of alcohol intake when completing questionnaires.
Habitual heavy alcohol drinking is associated with an increased risk of AF, although several study limitations exist and must be recognized. The relationship between daily alcohol consumption and the risk of AF was explained by a linear dose-response model, suggesting that not consuming alcohol at all is the most favorable behavior for avoiding AF rather than moderate alcohol consumption. Further investigation is needed to establish the extent to which this association is explained by a causal relationship.
Drs. Sone and Kodama are recipients of a Grant-in-Aid for Scientific Research (20300227) and Postdoctoral Research Fellowship (202965), respectively, both from the Japan Society for the Promotion of Science. This work is also financially supported by the Japan Cardiovascular Research Foundation and Ministry of Health, Labor, and Welfare, Japan. The sponsors had no influence over the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, or approval of the manuscript. The authors have reported that they have no relationships to disclose.
- Abbreviations and Acronyms
- atrial fibrillation
- confidence interval
- odds ratio
- paroxysmal atrial fibrillation
- relative risk
- Received May 18, 2010.
- Revision received July 26, 2010.
- Accepted August 10, 2010.
- American College of Cardiology Foundation
- Schoonderwoerd B.A.,
- Smit M.D.,
- Pen L.,
- Van Gelder I.C.
- Balbao C.E.,
- de Paola A.A.,
- Fenelon G.
- Costanzo S.,
- Di Castelnuovo A.,
- Donati M.B.,
- Iacoviello L.,
- de Gaetano G.
- Mattioli A.V.,
- Bonatti S.,
- Zennaro M.,
- Mattioli G.
- National Institute of Alcohol Abuse and Alcoholism
- Egger M.,
- Davey Smith G.,
- Schneider M.,
- Minder C.
- Di Castelnuovo A.,
- Rotondo S.,
- Iacoviello L.,
- Donati M.B.,
- De Gaetano G.
- Mukamal K.J.,
- Tolstrup J.S.,
- Friberg J.,
- Jensen G.,
- Gronbaek M.
- Denison H.,
- Jern S.,
- Jagenburg R.,
- Wendestam C.,
- Wallerstedt S.
- Preedy V.R.,
- Siddiq T.,
- Why H.,
- Richardson P.J.
- Lazarevic A.M.,
- Nakatani S.,
- Neskovic A.N.,
- et al.
- Athyros V.G.,
- Liberopoulos E.N.,
- Mikhailidis D.P.,
- et al.